LECTURE 5: PROSPECT THEORY
(for practice questions on lecture 4, take a look at the first part of lecture 5)
Prescriptive vs. Descriptive:
- Prescriptive/normative theories: what we SHOULD do o Expected value
Used to maximize $$
Failed to account for diminishing marginal utility
o Expected utility (Bernoulli)
Used to maximize utility
we do not think of outcomes as states of wealth. Instead, we use relativity which is based on
changes as opposed to absolutes
- Descriptive theories: what we ACTUALLY do
Case 1: If someone gives you $1000 and tells you If you want to learn more check out What is the meaning of household in family life?
that you can either get $500 automatically or take
a 50/50 chance to win either $1000 or $0 more,
what would you choose? Most people would choose
to take the $500.
Case 2: If someone gives you $2000 and tells you
that you either have to lose $500 or take a 50/50
chance to lose either $1000 or $0, what would you
choose? Most people would choose to take the risk
The two cases above have the same wealth
change, but there is a natural difference in
response: Case 1: risk aversion preference. Case 2: We also discuss several other topics like What is entropy in systems theory?
risk seeking preference.
The Isolation Effect: when the current state is
ignored. What’s important is relative gains and
o Prospect Theory:
Prospect theory is a theory that describes how people decide between two alternatives under risk and it
corrects two of Bernoulli’s assumptions:
1. Carriers of utility are relative gains and losses as
opposed to states of wealth
2. Outcomes have weighted averages, but not exactly probabilities
Prospect Theory: V (G )=π( X)∗v (X) +…
o Where V is the value function (the s-curve)
Reference Point: Utility is not based on
not states of wealth, but based on relative
gains and losses, relative to a reference
point of a decision (usually status quo)
Loss Aversion (greater slope for losses):
Gains sand losses aren’t symmetric. Losses We also discuss several other topics like What is needed to gain a competitive advantage over competitors?
have greater utility than gains of the same
Diminishing marginal utility (makes an s
∙ Gains - has concave shape, predicts
∙ Losses – has convex shape, predicts
o Where π is the decision weight function:
Increasing chances from 99% to 100% or
increasing chances from 0% to 1% make a
bigger subjective difference than going from
47% to 48% because decision weights are
steep near certainty. Don't forget about the age old question of Do gases have weak intermolecular forces?
Low probabilities overweighed, high
Example: Lotteries: We also discuss several other topics like How can you determine the effects of intermolecular forces on the properties of matter?
∙ If the value function predicts risk
aversion for gains, why do we buy
lotteries? Because small probabilities
are overweighted (decision weight
Prospect Theory shows why we make irrational decisions: - Framing:
o The framing of a question can reverse what people’s responses depending on perceptions of gains and losses o Example: Asian Disease Problem – There are 60,000 people with a disease.
∙ Program A: 20,000 people saved Don't forget about the age old question of What is john maynard keynes best known for?
∙ Program B: 1/3 chance 60,000 are saved, and
2/3 chance no one is saved
∙ Program C: 40,000 people die
∙ Program D: 1/3 chance no one dies, and 2/3
chance 60,000 die
People will choose Program A over B and program
D over C because of risk averse for gains and risk
seeking for losses
- The Endowment Effect:
One person owns a mug and is going to sell it.
Another person is going to buy this mug. The seller
would give a much higher price for selling the mug
than the buyer would give for buying the mug
Consequence of loss aversion
- Status Quo Bias:
o A biased preference for the current state of things
o It’s a consequence of loss aversion. Since losses have more powerful utility than gains, change can be more difficult when the options have equal value
When it comes to cars, you care about GPS
presence and safety equally. You currently have a
car with a GPS but is not very safe. You have 2
options for a new car:
∙ Car 1: has a GPS, not very safe
∙ Car 2: does not have a GPS, very safe
You would choose car 2 because of loss aversion:
you are more scared of loss than you are in seek of
- Diminishing Marginal Utility:
Scenario 1: If you drive to another town, you can
get a refrigerator for $50 instead of $100, would
Scenario 2: If you drive to another town, you can
get a car for $49,900 instead of $50,000, would
You are more likely to drive in scenario 1, since
there is a larger subjective difference.
- Prospect Theory: V (G )=π( X)∗v (X) +…
o Value function:
Reference Point: Utility based on relative gains and
losses, relative to a reference point of a decision
Loss Aversion (greater slope for losses): Losses
have greater utility than gains of same degree
Diminishing marginal utility (makes an s-shape
curve, gains predict risk aversion (concave) losses
predict risk seeking (convex)
o Decision weight function:
Probabilities aren’t weighed linearly
Decision weights are steep near certainty
Low probabilities overweighed, medium-high
LECTURE 6: Attribution Substitution Heuristics
How to study decision-making:
- Bounded rationality – satisfice instead of maximize, we find good enough solutions by “filling in”/using heuristics
- Heuristics: a shortcut to getting satisfactory solutions quickly (rules of thumb)
- Since heuristics don’t follow normative rules they lead to systematic and predictable errors/biases, which reveal the heuristics we use and tell us how the brain works
- i.e. discover bias- identify heuristic
- Psychologists have been using this approach to study judgment and decision-making for 30 years.
Heuristics and Biases:
Attribute Substitution Heuristic:
- When asked question we’re not evolved to answer optimally - We don’t answer the exact question, but a slightly different question we know the answer to
- This will occur when 3 conditions satisfied
o target attribute is relatively inaccessible
o related attribute/value is more accessible
o we do not reject the substitution of heuristic attribute even though we are conscious of it
- Bias: Base rate neglect and the Conjunction Fallacy o How to identify the bias: we judge the probability of the conjunction of two events to be more than that of each of the elements alone, which is not logically possible
o Judging probability by similarity (stereotypes), ignoring
o Example: The Linda Problem
o We should combine diagnostics and base rates when making a conclusion, but we overweight diagnostics and this leads to bias
- Bias: Probability Overestimation
o How to identify this bias: Estimated probability compared to objective probabilities
o What is easier to call to mind is what we use to inform a decision, ease of retrieval as opposed to number of instances
o Example: First, estimate how many words in 2 pages of a book would have this pattern _ _ _ _ _ _ n _. Now estimate how many words would have this pattern _ _ _ _ _ i n g. People tend to estimate much less words for the first pattern even though it is broader than the second pattern.
o Example2: One group of people were asked to recall 6 examples of when they were assertive and rank their assertiveness. Another group of people were asked to recall 12 examples of when they were assertive and rank their assertiveness. The group who had to come up with 12 examples tended to rank themselves as less assertive b/c it became harder for them to recall examples. The greater psychological struggle in the act of recalling made them think they were less assertive in reality.
- Bias: the duration of an event is not important when it comes to judging goodness/badness
Heuristic: Peak-End Heuristic:
o We confuse the experiencing self with the remembering self
o We perceive happiness by averaging the peak and end feelings
o Example: Daniel Kahneman’s colonoscopy experiment - people who suffered less thought they suffered more because their treatment stopped at high pain while the others had the pain decrease before stopping treatment, even though they suffered the same peak at one point
- Bias: Scope neglect
o We ignore numerical values in problem descriptions Heuristic: Affect Heuristic:
o Assign values by feeling, as opposed to number
o Example: How much would you pay to save 2,000 birds? How much would you pay to save 20,000 birds? How
much would you pay to save 200,000 birds? Not only
were answers not proportional, but the average price for 2,000 birds was greater than that for 20,000 birds.
o Example2: Jury, compensating for loss: it’s very
subjective, we don’t know how to compensate for
physical and mental pain
- Bias: Effort Bias
o Effort/time spent translates to how valuable something is - Heuristic: Effort Heuristic:
o When something is easy to judge, effort is not a factor considered
o When something is hard to judge, effort becomes an important factor in judgment because of attribute
substitution (we don’t know how to evaluate the actual thing, so we find a related attribute to assess instead) o Example: We want to buy a painting, but we don’t know how much we’d pay for it. We would pay more money for it if there was relatively more time spent on it.
Overview of Attribute Substitution:
We assess attributes that are easier to evaluate:
- Instead of probability, evaluate:
o Representativeness (similarity)
o Availability (ease of retrieval)
- Instead of feelings over time, evaluate:
o Peak-end heuristic (crucial moments)
- Instead of numerical value, evaluate:
o Affect heuristic (feeling)
- Instead of quality, evaluate:
o Effort heuristic (effort)
Attribute substitution heuristics help relieve the effort required in decision-making:
- Examine less cues
- Facilitate retrieving/storing cue values
- Simplify assessing weights for cues
- Integrate less info
- Assess fewer alternatives
- We do not seek optimality in our reasoning and rationality (bounded rationality)
- Reasoning is driven by heuristics
o Heuristics allow us to live efficiently, our intuitions are often right
o But, heuristics also lead to biases
NOTES on reading Chapter 3: Judging Probability and Frequency
- Extension of the availability heuristic
- Support theory says that people judge hypotheses as opposed to events by considering the either implicit or explicit support for the hypotheses
- Example: When students were asked to judge the probability of death due to natural causes (which is an “implicit
disjunction” in this case because it can be split into more specific categories), they said 58%, but when students were asked to give probabilities for death due to heart disease, cancer, and other natural causes, the probability was 73% for other natural causes.
o Therefore, for an implicit disjunction, support is usually subadditive b/c support is relatively less compared
explicit disjunctions (if the implicit disjunction was split into more specific components)
Minerva-DM: A memory model of judgment:
- The heuristics approach to describing processes of judgment is lacking in that it cannot give a good picture of memory and judgment processes integrated together
- MDM is a multiple-trace memory model that tries explain representativeness and availability broadly in terms of memory. It says a memory trace is created for each event experienced (instead of updating). A memory probe is created to access a memory and when a trace has a weak similarity to the probe, then the “echo” is weak and vise versa. As a result, this theory proposes that when we make judgments of likelihood, we use the “echo intensity” from the memory traces of MDM.
- Example: In tackling the Linda problem with MDM, we would create memory probes for the information given on Linda. When judging the probability for Linda is a bank teller, there is not as much echo intensity for bank teller b/c the memory probes created seem to match feminist more, but when we judge the probability of the conjunction that Linda is a feminist and bank teller, there is more echo intensity.
- In other words, a conjunctive probe will give a higher echo intensity for A∩ B than for A and B alone because it will
combine the echo intensities of both A and B.
- Therefore, bias can arise from the strength of encoding memories
- We don’t know how accurate MDM is in modeling memory, so we don’t know how capable it is in explaining judgment phenomena.
- There are other theories that say highly distinct memories are encoded as separate memory traces and similar experiences are assimilated.
- Remember that there are other theories for how we make judgments in addition to heuristics and MDM.