Bounded Rationality, Intuition, and Decision Making: Dual-Systems Theory, Accessibility, Prospect Theory, and Prototype Heuristics

Foundations of Bounded Rationality and Dual-System Theory

• The study of intuitive judgment and decision making under uncertainty operates within the framework of bounded rationality, a concept defined and named by Herbert A. Simon.

• Intuitions are defined as thoughts and preferences that come to mind quickly, automatically, and without detailed reflection or deliberate computation.

• Cognitive operations are partitioned into two distinct modes of function, designated by Keith Stanovich and Richard West as System 1 and System 2:

  • System 1 (Intuitive System): Operations are fast, automatic, effortless, associative, implicit (unaccessible to introspection), emotionally charged, governed by habit, and difficult to control or modify.

  • System 2 (Controlled System): Operations are slower, serial, effortful, consciously monitored, deliberately controlled, flexible, and potentially rule-governed.

• Cognitive Load and Dual-Task Interference:

  • Total capacity for mental effort is limited. Effortful processes (System 2) interfere with one another, whereas effortless processes (System 1) neither cause nor suffer significant interference when combined with concurrent tasks.

  • Secondary cognitive tasks (e.g., holding a sequence of digits in memory) deplete System 2 resources, increasing the probability that System 1 impressions are expressed directly without filtering or self-monitoring.

• Percepts vs. Conceptual Impressions:

  • System 1 and the perceptual system automatically generate non-voluntary, verbally implicit impressions of the attributes of perceived objects and thoughts.

  • System 2 converts these impressions into explicit, intentional judgments. An intuitive judgment directly reflects an un-modified System 1 impression.

• Cognitive Self-Monitoring and Lax Control:

  • System 2 serves a monitoring function over System 1 outputs and overt behavior, but this supervision is normally lax and permissive.

  • The Bat and Ball Problem: "A bat and a ball cost $1.10 in total. The bat costs $1.00 more than the ball. How much does the ball cost?"

  • Intuitive answer (System 1): 10 cents ($10\text{ cents}$).

  • Correct answer (System 2): 5 cents ($5\text{ cents}$) (since $1.05 + $0.05 = $1.10).

  • Error rates: $50\%$ ($47/93$) of Princeton students and $56\%$ ($164/293$) of University of Michigan students yielded to the immediate intuitive response without self-checking.

  • Errors on such simple puzzle tasks serve as significant predictors of impatience/intolerance of delay and cheating behavior.

• Skilled Intuition:

  • System 1 operations are not inherently flawed; prolonged practice yields exceptional intuitive skill where complex computations occur rapidly and effortlessly.

  • Examples include chess grandmasters observing a board and instantly identifying "White mates in three," or experienced nurses detecting subtle signs of impending heart failure before explicit clinical test results emerge.

  • Decision makers can perform superiorly when trusting trained intuitions rather than engaging in detailed, explicit analysis.

Accessibility and Natural Assessments

• The Accessibility Continuum:

  • Accessibility: The ease or effort with which particular mental contents come to mind.

  • Accessibility is a continuous spectrum ranging from rapid, automatic perceptual processes to slow, serial, effortful deliberate calculations.

  • Determinants include stimulus salience, selective attention, specific training, associative activation, emotional arousal, and contextual priming.

• Selective Accessibility Examples:

  • In a vertical tower of blocks, height, top area, and volume are immediately accessible as perceptual impressions, whereas the total surface area covered if dismantled is inaccessible without deliberate computation.

  • When blocks are spread flat on a surface, total area is immediately accessible, while the total height of a potential constructed tower is inaccessible.

• Ensemble Statistical Properties:

  • When presented with a set of similar objects (e.g., an array of lines), a prototypical representation (such as the average length) is computed automatically as an accessible perceptual impression.

  • Extensional ensemble properties (such as the total sum of line lengths) are inaccessible and require explicit System 2 summation.

• Natural Assessments:

  • Attributes routinely and automatically registered by System 1 without conscious intention or effort.

  • Includes physical dimensions (size, distance, loudness), abstract relational properties (similarity, causal propensity, surprisingness), affective valence (good/bad evaluation), mood, and cognitive fluency (accessibility itself).

  • Primordial evaluation: Neural circuitry (e.g., amygdala pathways) rapidly evaluates stimuli as good (approach) or bad (avoid) before explicit conscious identification.

• Context and Ambiguity Suppression:

  • System 1 automatically suppresses visual and conceptual ambiguity, outputting a single contextually plausible interpretation without subjective awareness of alternative readings.

• An identical ambiguous symbol is perceived as the letter "B" in the context of "A B C" and as the number "13" in the context of "12 13 14".

• Doubt and subjective uncertainty are meta-cognitive functions of System 2, requiring the deliberate holding of mutually incompatible interpretations in mind.

Framing Effects and Invariance Violations

• Extensionality and Invariance:

  • Invariance (Extensionality): An essential axiom of standard economic rationality stating that choices and preferences should not be altered by superficial or logically equivalent variations in the description of options or outcomes.

  • Human choices systematically violate invariance due to differential salience and accessibility induced by framing.

• The Asian Disease Problem:

  • An unexpected disease outbreak is estimated to kill $600\text{ people}$.

  • Gain Frame:

  • Program A: $200\text{ people}$ will be saved. [Chosen by a substantial majority; demonstrates risk aversion].

  • Program B: $\frac{1}{3}$ probability that $600\text{ people}$ will be saved, and $\frac{2}{3}$ probability that no people will be saved.

  • Loss Frame:

  • Program A' $400\text{ people}$ will die.

  • Program B' $\frac{1}{3}$ probability that nobody will die, and $\frac{2}{3}$ probability that $600\text{ people}$ will die. [Chosen by a clear majority; demonstrates risk seeking].

  • Logical Equivalence: Saving $200\text{ of } 600$ is identical to $400\text{ dying}$. Certainty of saving lives is disproportionately attractive; certainty of deaths is disproportionately aversive.

• Medical Treatment Preferences:

  • Options presented as $\text{survival rates}$ ($90\%$ short-term survival) vs. $\text{mortality rates}$ ($10\%$ immediate mortality).

  • The survival frame generates a significantly higher preference for surgery over radiation therapy among both lay patients and experienced physicians.

• Child Custody Decisions (Choose vs. Reject Frame):

  • Parent A: Average income, average health, average working hours, reasonable relationship with child, relatively stable social life.

  • Parent B (Rich description): Above-average income, very close relationship with child, extremely active social life, lots of work-related travel, minor health problems.

  • Results: Parent B is chosen by a majority when the task is framed as "Award Custody" (salient positive attributes), and Parent B is ALSO chosen by a majority when framed as "Deny Custody" (salient negative attributes).

• Problem Isomorphs:

  • Transformation puzzles logically identical to the Tower of Hanoi exhibit drastically different solution rates based on representation (e.g., monsters passing balls vs. monsters changing ball colors).

  • Passive acceptance: Decision makers passively accept the frame provided rather than spontaneously translating problems into canonical canonical representations.

Prospect Theory and Reference Dependence

• Bernoulli's Error:

  • Daniel Bernoulli ($1738$) posited that psychological value (utility) of gambles is evaluated based on absolute states of final wealth ($W$), assuming a logarithmic utility function.

  • Bernoulli's Error: Expecting utility to be reference-independent. Human perceptual and cognitive systems do not evaluate absolute magnitudes; they evaluate changes relative to an adaptation level or reference point.

• Visual analogy: Two square patches of identical physical luminance appear completely different in brightness depending on whether the surrounding field is dark or light.

  • Demonstrations of Reference Dependence:

• Problem 2: Accept a gamble with a $50\%$ chance to win $\$150$ and a $50\%$ chance to lose $\$100$? Most reject this gamble unless the gain is $2\text{ to } 2.5\times$ larger than the loss. A shift in overall wealth of $-\$\!100$ does not alter choice.

• Problem 3: Lose $\$100$ with certainty OR take a $50\%$ chance to win $\$50$ and a $50\%$ chance to lose $\$200$? Most select the gamble (risk seeking). A shift in overall wealth of $+\$\!100$ does not alter choice.

• Problem 4: Wealth $W$ vs. $50\%$ chance of $W + \$100$ and $50\%$ chance of $W - \$150$. When framed as absolute wealth $W$, people exhibit mild preference for the gamble, treating stakes as negligible.

• Problem 5: Person A's wealth goes from $4\text{M}$ to $3\text{M}$. Person B's wealth goes from $1\text{M}$ to $1.1\text{M}$. Bernoulli evaluates Person A as having higher utility (higher absolute wealth state). Prospect theory evaluates Person B as happier today because value attaches to transitions and short-term changes.

  • The Value Function of Prospect Theory:

• Defined explicitly on gains and losses relative to a neutral reference point ($0$).

• Concave in Gains: Produces risk aversion for positive outcomes.

• Convex in Losses: Produces risk seeking for negative outcomes.

• Loss Aversion: The function is sharply kinked at the reference point; losses loom larger than gains by a ratio of approximately $2.0\text{ to } 2.5$.

• Approximated by power functions in both domains with exponents less than unity: $v(x) = x^\alpha$ for gains and $v(x) = -\lambda(-x)^\beta$ for losses (where $\alpha, \beta < 1$).

  • Applications of Loss Aversion:

• Endowment Effect: Minimal selling prices (WTA) for an owned object exceed maximum buying prices (WTP) by a factor of $2$ or more.

• Status-Quo Bias: The current position acts as the reference point; potential disadvantages of changing positions loom significantly larger than potential advantages.

• Out-of-Pocket vs. Foregone Gains: Direct financial expenditures are evaluated as explicit losses, whereas foregone cost reductions are evaluated as unearned gains, impacting legal rulings and consumer notions of market fairness.

  • Narrow Framing and Mental Accounting:

• Rational agency assumes broad framing—evaluating choices in a comprehensively inclusive context considering all overall risks and lifetime opportunities.

• Human behavior relies on narrow framing: decisions are evaluated in isolation as single, localized problems.

• Aggregating repeated gambles across time shifts preference toward risk neutrality, yet individuals routinely treat individual decisions as isolated events unless explicitly forced to aggregate them.

Attribute Substitution Model of Judgment Heuristics

• Mechanism of Attribute Substitution:

  • Occurs when an individual is tasked with evaluating a difficult target attribute of an object and involuntarily substitutes a highly accessible heuristic attribute instead.

  • The judge maps the assessment of the heuristic attribute directly onto the scale of the target attribute without recognizing the substitution.

• Perceptual Analogy: Estimating the 2D height of two printed horses on a page with perspective background cues. The perceptual system automatically substitutes 3D spatial size (heuristic attribute) for 2D page image size (target attribute), producing a powerful visual illusion.

  • Empirical Validation of Attribute Substitution:

• The Tom W. Problem:

  • Tom W. is described with stereotypical traits: neat, mechanical, lacking creativity, low empathy, sci-fi interest.

  • Base-rate group estimated student enrollment percentages across $9\text{ fields}$ (e.g., Humanities $20\%$, Library Science $3\%$).

  • Similarity group ranked how much Tom W. resembled a typical student in each field.

  • Probability group ranked the likelihood that Tom W. is enrolled in each field (noting description was based on unvalidated high school tests).

  • Results: Correlation between Probability rankings and Similarity rankings was $r = 0.98$. Correlation between Probability rankings and Base-Rates was $r = -0.63$. Respondents substituted similarity for probability completely, demonstrating extreme base-rate neglect.

• The Linda Problem and the Conjunction Fallacy:

  • Description: Linda is $31\text{ years old}$, single, outspoken, bright, philosophy major, concerned with discrimination and social justice.

  • Rank probability/similarity of outcomes, including: "Linda is a bank teller" ($T$) vs. "Linda is a bank teller and active in the feminist movement" ($T \& F$).

  • $85\%$ of similarity group ranked $T \& F$ higher than $T$.

  • $89\%$ of probability group ranked $T \& F$ higher than $T$.

  • Conjunction Fallacy: Probability of a conjunction ($P(A \cap B)$) cannot exceed the probability of its constituents ($P(A)$). Probability rank matched similarity rank ($r = 0.98$), demonstrating complete substitution of representativeness for probability.

• Survey Order Effects (Dating and Happiness):

  • Questions: "How happy are you with your life in general?" and "How many dates did you have last month?"

  • When general happiness was asked first: correlation was negligible ($r \approx 0$).

  • When dating frequency was asked first: correlation skyrocketed to r = 0.66$.\n * Dating satisfaction became the accessible heuristic attribute substituted for overall life satisfaction.\n* **The Affect Heuristic:**\n * Primary affective reactions (good/bad, like/dislike) act as general-purpose heuristic attributes substituted for complex evaluations.\n * Evaluating public causes, regulatory risk level, or benefit-cost ratios: Willingness to pay, perceived danger, and moral satisfaction all map back to a single underlying affective evaluation.\n * Jurors' punitive damage awards reflect the direct mapping of experienced outrage onto a dollar scale.\n\n\n# Prototype Heuristics and Nonextensional Thinking\n\n* **Extensional vs. Prototype Attributes:**\n * **Extensional Attribute:** An aggregate value over a set or category that obeys conditional adding and monotonicity (e.g., total volume, total surface area, total probability, total financial value, total suffering).\n * **Prototype Attribute:** A measure of central tendency or average characteristic of a set's prototype (e.g., average line length, category similarity, typical pain intensity).\n * **Prototype Heuristic:** Substituting a prototype attribute (an average) for an extensional attribute (a sum or aggregate).\n* **Biases Produced by Prototype Heuristics:**\n 1. **Extension Neglect (Scope / Duration Neglect):** Insensitivity to variations in set extension (size, duration, count).\n 2. **Violations of Monotonicity:** Adding non-zero elements to a set decreases its judged value if the added elements lower the set average.\n* **Empirical Studies of Extension Neglect:**\n * *Willingness to Pay (WTP) for Public Goods:* Desvousges et al. (1993) measured WTP to save migratory birds from drowning in oil ponds.\n * 2!,000\text{ birds: } \$80\n * 20!,000\text{ birds: } \$78\n * 200!,000\text{ birds: } \$88\n * Estimated amounts were virtually flat despite a 100\times increase in bird scope. Respondents evaluated the prototype image of a single distressed bird.\n * *Duration Neglect in Medical Experiences:* Redelmeier and Kahneman (1996$) recorded real-time pain during colonoscopy procedures every$60\text{ seconds}.\n\n\n\n * Patient A: Procedure length 8\text{ minutes}$, ended at peak pain intensity ($8/10).\n * Patient B: Procedure length 24\text{ minutes}$, ended at low pain intensity ($1\text{ to } 3/10).\n * Global retrospective evaluation correlated strongly with the **Peak-End Rule** (average of peak pain and end pain, r = 0.67$) but showed zero correlation with total duration ($r = 0.03). Patient A remembered the procedure as significantly worse than Patient B.\n* **Empirical Studies of Monotonicity Violations:**\n * *Dinnerware Evaluation (Hsee, 1998):*\n * Set A (24\text{ pieces}): All intact.\n * Set B (40\text{ pieces}$): The same$24\text{ intact pieces}$+$8\text{ intact pieces}$+$8\text{ broken pieces}$($32\text{ intact}, 8\text{ broken} total).\n * **Separate Evaluation:** Mean WTP was \$33$for Set A ($24\text{ pieces}$) and$\$23$for Set B ($40\text{ pieces}). Adding intact and broken pieces degraded the average quality of the set, causing a lower price.\n * **Joint Evaluation:** When both sets were evaluated side-by-side, respondents showed choice by dominance, pricing Set B higher than Set A.\n * *Urn Sample Evidence (Griffin & Tversky, 1992):*\n * Urn 1 has 70\%$Red,$30\%$White. Urn 2 has$30\%$Red,$70\% White.\n * Sample 1 (3\text{ Red}, 0\text{ White}$) yields a prototype proportion of$100\% Red.\n * Sample 2 (7\text{ Red}, 3\text{ White}$) yields a prototype proportion of$70\% Red.\n * Bayesian normative support depends on difference (R - W$). Sample 2 offers stronger support ($7 - 3 = 4$) than Sample 1 ($3 - 0 = 3).\n * Intuitive judgment relies on proportion (prototype attribute): people express far higher confidence in Sample 1 (3\text{R}, 0\text{W}$) than Sample 2 ($7\text{R}, 3\text{W}).\n * *Cold-Pressor Experiment (Kahneman et al., 1993):*\n * Short trial: Hand immersed in 14\,^\circ\text{C}$water for$60\text{ seconds}.\n * Long trial: Hand immersed in 14\,^\circ\text{C}$water for$60\text{ seconds}$, plus an additional$30\text{ seconds}$where temperature rose gradually to$15\,^\circ\text{C}.\n * A clear majority of participants deliberately chose to repeat the **long trial**, preferring extra total pain because the end period was less intense, satisfying the Peak-End prototype rule.\n\n\n# System 2 Interventions and Debias Conditions\n\n* **Corrective Operations:**\n * Errors of intuitive judgment require failures of both systems: System 1 generates the erroneous intuition, and System 2 fails to detect and override it.\n * System 2 intervention depends on the relative accessibility of corrective rules, statistical principles, or metacognitive cues.\n* **Factors Impairing System 2 Corrections:**\n * Time pressure.\n * Concurrent cognitive load / secondary mental tasks.\n * Circadian mismatch (e.g., morning people tested in the evening).\n * Positive mood / happy affect (promotes mindless reliance on System 1 scripts).\n* **Factors Enhancing System 2 Corrections:**\n * High cognitive ability / intelligence.\n * High Need for Cognition score.\n * Formal exposure to statistical reasoning and training in the logic of sets.\n * Explicit reminders and domain cues (e.g., instructing subjects to "think like a statistician").\n* **Metacognitive Discounting:**\n * Schwarz & Clore (1983$$): Ambient rainy weather lowers self-reported life satisfaction. Asking respondents "How is the weather over there?" immediately eliminates the mood effect by making subjects metacognitively aware of the irrelevant bias source.

• Oppenheimer: Awareness of surname fame eliminates availability bias; forced writing of initials reverses availability effects in frequency judgments.

  • Five Modes of Judgment and Choice Output:

• An intuitive response is generated by System 1, and:

  • (a) Endorsed directly by System 2.

  • (b) Adjusted (insufficiently) for other features recognized as relevant.

  • (c) Corrected (or overcorrected) due to explicit recognition of a bias.

  • (d) Blocked completely from overt expression because it violates a valid rule.

• No intuitive response comes to mind; judgment is computed from scratch by deliberate System 2 operations.

  • Frequency of Outcomes: Cases 1(a) and 1(b) are the overwhelmingly predominant modes of human cognitive operation. Case 1(d) is extremely rare.