Probability Judgement Flashcards
Normative, Descriptive, Prescriptive
Normative is how people should make decisions according to theory of rationality.
Descriptive is the way people actually tend to behave
Prescriptive how we might use knowledge to help people make better decisions
Coherence
Internal consistency between your judgements
Correspondence
Accuracy with respect to world
Money Pump Problem
Thought experiment, shows preferences can be intransitive - not coherent.
If I would pay more ££ for A then B, and Pay more ££ for B then C, BUT would pay more for C then A - could theoretically create a ‘money pump’
Expected Value
Expected Value =(O1 * P1) + (O2 * P2)
St Peters Paradox Paradox
Shows failure of expected Value to capture how people actually make decisions
Toss coin, every time lands on tails, double prize - starting with £2 - How much would you be willing to pay to play?
Paradoxical as has infinite expected value
Expected Utility
Utility doesn’t equal value
For example the utility of money decreases as £ increases
Bounded Rationality Theory
How people actually make decisions
Based on imperfect knowledge
Context sensitivity
“satisficing” - adequate rather than optimal
Herbert Simon - economic decision making
Alias Paradox
Choice problem showing inconsistency of observed choices with predictions of expected utility
In first gamble less risky choice is preferred over a higher expected utility, while in the second gamble a higher expected utility is preferred over a less risky choice.
Brunswick Lens Model
Indirect interaction with objects / events
Internal perceptions of external events mediated through sense organs
“Lense of Cues”
Multiple Cue Probability Learning
Experimental Paradigm for learning relationships between cues and events
- Learning over many trials, hypothesis testing
- Learning to to Predict outcome
- Cues in environment vary in validity
Key processes in making judgments
1 - Discovering Information
2 - Acquiring & searching through Information
3 - Combining information
4 - Feedback
1 - Discovering Information
- Identification of new valid predictive cues
- Incorporation into mental model
Klayman (1988)
2 - Acquiring Information
*When information is costly DM needs to balance against improved Decision accuracy
People tend to use under optimal samples, why? costs, working memory, amplification effect?
2 - Searching for Information
- Alternative based strategy - consider each option in turn
* Consider by attribute (easier elimination)
Elimination by Aspects Model (EBA)
A theory of choice - presented by Tversky (1972)
Identify a dimension, alternatives that don’t posses this are eliminated
Choose aspects acording to weight
Compensatory versus non-compensatory stratagies
Compensatory strategies allow trade off between attributes, ie a high score in a lower weighted attribute can compensate for low score in a more important attribute.
Conflict confronting, e.g Weighted Additive strategies.
** Linear Model as all info included
Non compensatory strategies do not allow trade off - e.g if most important attribute scores low, that alternative will not be chosen - eg lexicographic. (Conflict avoiding)
3 - Combining Information
How process information to make decision?
Include all cues? Weight information same?
Compensatory or non-compensatory?
Proper Linear models versus bootstrap model
Proper linear models -> optimized mathematically
Bootstrap model -> Formalise judgers use of cues, express judgment policy in model
Should wherever possible a human judge be replaced with a machine?
- Statistical models routinely outperform humans
- Statistical model always arrive at same judgment given a data set
- Computers process all of the data
- Humans prone to bias (confirmation, availability)
- BUT
- humans good at identifying & quantifying cues
- still need experts to provide information
Recognition Heuristic
If one of two things recognised and not the other - infer the recognized one have a higher value with respect to criterion
Maximising Expected Utility
Assessing decisions on states, outcomes, acts
EU = S Utility (outcome ij) * Probability (statej)
Four axioms of MEU Theory
Four axioms of expected Utility
1 - cancellation
2 - transitivity
3 - Dominance
4 - Invariance
1) Cancellation Axiom
If states of world give same outcome - regardless of choice, can eliminate from decision
E.g broken by Ellsberg paradox
2) Transitive axiom
Preferences should be transitive
LInked to money pump
3) Dominance
If A is better than B is 1 respect and at least as good as in all other - A should always be preferred
4) Invariance
Given the same information we should make same decisions regardless of how presented
Prospect Theory
People evaluate gambles in terms of losses of gains according to a reference point
Concave function of value - losses loom larger than gains
Reference point open to manipulation
