AI summary by TwinMind
Unit 10: Utility Analysis and Multidimensional Evaluation, Video 3: Conditions for Value Function
MIT OpenCourseWare
9m 43s
Introduction to Value Function Conditions
- Value function construct requires three fundamental assumptions/axioms
- These conditions determine when legitimate, meaningful, unambiguous value functions are possible
- Important to recognize when value functions cannot exist
- Large optimization/valuation literature fails to acknowledge these limitations
First Assumption: Completeness (Complete Pre-order)
- People have preferences over all choices
- Doesn't require conscious consideration of all possible options
- Must be capable of making preference decisions on any choice
- Permits equal preference statements (e.g., chocolate vs strawberry ice cream having same value)
- Indifference is acceptable - don't need strict preference ordering
- Logical necessity - without preference capability, no value function possible
Second Assumption: Transitivity
- If X1 preferred to X2, and X2 preferred to X3, then X1 preferred to X3
- Generally assumed true for individuals
- Context-dependent phrasing can create apparent violations
- Example: strawberry > chocolate > marshmallow → strawberry > marshmallow
- Critical limitation: Not necessarily true for groups
- Board of directors decisions
- Design teams
- City councils
- Business groups/factions
- Real example: Morocco company with conflicting departmental values
- Finance people vs marketing people vs industrial people
- Different value sets for spending decisions
- No consistent group value function
Student Discussion: Transitivity Exceptions
- Reference to Savage's axioms for subjective expected utility
- Ellsberg paradox demonstrates transitivity failures under ambiguity
- People prefer non-ambiguous options
- Violates transitivity under uncertainty
- Allais paradox also mentioned
- Psychological experiments show context effects on preference ordering
- Acceptable for individual decisions but problematic for groups
Third Assumption: Archimedean Principle (Monotonicity)
- Values between best and worst outcomes are weighted functions
- Value of intermediate outcome = proportion of best + complementary proportion of worst
- Requires monotonic preferences (always increasing or always decreasing)
- No up-down-up patterns in valuation
- Necessary for transitivity and consistent value functions
Monotonicity Assumption Challenges
- Not always reasonable or true in practice
- Examples of non-monotonic preferences:
- Speaker volume: want audible but not too loud
- Salt in food: better up to point, then worse
- Sugar in coffee: optimal level exists
- Workarounds possible in some cases:
- Availability vs usage distinction (salt on table vs using it)
- Benefits level off rather than become negative
- Monotonicity assumption often doesn't apply without creative problem restructuring
