reflections
NUm
02-UncertaintyRisksandHeuristics1.pdf
BUS143 Topic 2
1
Uncertainty, Risks, and Heuristics
BUS143: Judgment and Decision Making
Ye Li All rights reserved ®
Please download Moblab if you don’t have it already!
How do we form these judgments?
1. In all low-income countries across the world today, how many girls finish primary school? • 20%, 40%, or 60%
2. Where does the majority of the world population live? • Low, Middle or High countries
3. In the last 20 years the proportion of the world population living in extreme poverty has? • Almost doubled, Remained more or less the same, Almost
halved
4. What is the average life expectancy in the world today? • 50, 60, or 70 years
5. There are 2 billion children in the world today aged 0-15 years old, how many children will there be in 2100 according to the UN? • 4 billion, 3 billion, or 2 billion
6. The UN predicts that by 2100 the world population will have increased by another 4 billion people, what is the main reason? • There will be more children aged below 15 • There will be more adults aged 15-74 • There will be more very old people aged 75 and older
7. How did the number of deaths per year from natural disasters change over the last 100 years? • More than doubled, Remained about the same, or
Decreased to less than half
8. There are about 7 billion people in the world today, approximately where do they live? • 1 billion in Europe, 4 in Asia, 1 in Africa and 1 in Americas • 1 billion in Europe, 3 in Asia, 2 in Africa and 1 in Americas • 1 billion in Europe, 3 in Asia, 1 in Africa and 2 in Americas?
9. How many of the world's 1 year old children today have been vaccinated against some diseases? • 20%, 50%, or 80%
10. Worldwide, 30 year old men have spent 10 years in school on average. How many years have women of the same age spent in school? • 9 years, 6 years, or 3 years
11. In 1996 tigers, giant pandas, and Black Rhinos were all endangered. How many of these species are critically endangered today? • 2 of them, 1 of them, or none of them
12. How many people in the world have some access to electricity? • 20%, 50%, 80%
13. Global climate experts believe that over the next 100 years the average temperature will on average...? • get warmer, remain the same, or get colder
BUS143 Topic 2
2
Decisions require uncertainty judgments
• Uncertainty: Uncontrollable events that decision-makers do not have total information about.
• Probability: Quantified beliefs about uncertain events. • How is uncertainty different from risk? – A risk has a known probability distribution. E.g., coin flip
• Who uses probability estimates in making choices? • Nearly every business (and other) decision involves some
estimate of likelihood – Alternatives – Beliefs
What if the probabilities are systematically biased?
– Consequences
Charlie Munger
“If you don’t get this elementary, but mildly unnatural mathematics of probability into your repertoire, then you go through a long life like a one-legged man in an ass-kicking contest.
One of the advantages of a fellow like Buffett, whom I’ve worked with all these years, is that he automatically thinks in terms of decision trees and the elementary math of permutations and combinations...”
Address to USC Marshall Business School
BUS143 Topic 2
3
Why should we quantify uncertainty?
• MobLab: What probability would you assign to the following verbal probability statements? (0 to 100%)
–“Usually” _____%
–“Possible” _____%
–“Somewhat likely” _____%
–“Probably” _____%
–“Fairly unlikely” _____%
Normative: Econs
Subjective expected utility =
Value(outcome) × Probability(outcome)
How do Econs use probabilities? – Stated probabilities: 20% = 20% – Estimated probabilities: Bayes’ Rule
BUS143 Topic 2
4
Bayes’ Rule Primer
Suppose that your friend has been feeling quite sick and thinks he or she has the new swine flu going around. Fortunately, there is a new quick diagnosis test for swine flu available. This test will make a positive diagnosis if you have swine flu 99.99% of the time. Your friend gets tested and it comes back positive… • What is the probability that your friend has the
swine flu? • What additional info do you need to know? – If you do not have swine flu, there is still a 1% chance that
the test will be positive (false positive). – 1% of Americans have swine flu
Flu diagnosis: Normative Analysis (blank)
Pr(Positive)=
Pr(Flu|Positive)= Pr(Positive|Flu)∙Pr(Flu)/Pr(Positive)
Normative answer depends on:
The base rate (1%)
Ex. 1: Pr(Flu) = 0.1% Pr(Flu|Positive) =
Ex. 2: Pr(Flu) = 10% Pr(Flu|Positive) =
The quality of the information (1% false positive rate)
Ex. 3: Pr(Positive|No Flu) = 10% Pr(Flu|Positive)=
Ex. 4: Pr(Positive|No Flu) = .01% Pr(Flu|Positive)=
1%
No Flu
99%
Positive|Flu
Negative|Flu
Flu
Positive|No Flu
Negative|No Flu
BUS143 Topic 2
5
Descriptive: Humans
• How do humans actually think about risk and uncertainty? – How do we actually use stated probabilities? (risk) – How do we estimate probabilities that we do not
know? (uncertainty)
• Answer: People use heuristics – Heuristics can lead to biases
Homo sapiens: Probability distortions
• Very small probabilities (e.g., 1/10,000) treated as larger than they actually are (e.g., 1/100)
• Almost certain events (e.g., 99%) treated as less certain than they actually are (e.g., 85%)
• 0% and 100% are special cases
BUS143 Topic 2
6
Homo sapiens: Probability Estimation
• Definition: Heuristics – Rules of thumb (shortcuts) that simplify judgments
and decisions – System 1
• Definition: Biases – When judgments and decisions deviate systematically
from what is considered optimal or appropriate – Sometimes caused by usage of heuristics
Major Heuristics under Uncertainty
1. Availability (judging by familiarity)
2. Representativeness (judging by resemblance)
3. Anchoring (judging from starting values)
• Heuristics can lead to overconfidence, which we’ll discuss next week
BUS143 Topic 2
7
Availability Example 1
The following 10 corporations were ranked by Fortune magazine to be among the 150 largest United-States-based firms according to revenue for 2019:
Group A: Starbucks, McDonald’s, Facebook, American Express, Nike
Group B: Kroger, Fannie Mae, United Health, McKesson, Amerisource Bergen
Compare these two groups to each other in terms of revenue for 2018:
Your estimate: A =________ / B = _________ REMINDER: You should not feel the need to Google answers for web assignments…
Cause of Death
Median Estimate (x000)
Actual (2018; x000)
Percent (Est./Act.)
Google News Search (2018)
Fire 5 2.60 192% 27,900,000
Lightning 100 total 26 total 384% 309,000
Motor vehicle accident 80 37.9 211% 11,400,000
Falls 5 33.0 15% 433,000
Homicide (murder) 40 15.8 253% 29,700,000
Suicide 50 42.8 122% 6,970,000
Terrorism 5000 total 80 total 6250% 3,860,000
Lung Cancer 75 155.6 48% 217,000
Breast Cancer 50 41.7 120% 2,340,000
Heart Disease 100 614.3 16% 942,000
Alzheimer’s Disease 50 93.5 53% 225,000
Availability Example 2: Reasoning by Recall
Estimate the number of people living in the US who die annually from each of the following causes. ~2.5 million deaths in US / year
BUS143 Topic 2
8
Availability
In forming a judgment, we tend to… • Make predictions and evaluations based on the ease with
which objects or instances come to mind – Rely heavily on readily available (recent, salient, vivid) information – Fail to discount the quality of the information – Fail to consider other possible sources of information
• More simply… – Things that are easy to “picture” are overestimated – Things that are hard to “picture” are underestimated
Sources of Availability
• What creates easy recall? – Frequent exposure Media, word-of-mouth, day-to-day experience
– Recent exposure (e.g., Flu, shootings in the news)
– Linking to what we already know Familiarity!
– Vividness Concrete, detailed events (not abstract descriptions) Emotional impact
BUS143 Topic 2
9
Implications for Consumer Demand
Imagine you are about to take a 1-week trip to Malaysia (Israel) as part of your UCR education. You do not have any insurance for this trip. No insurance is provided by the credit card company through which the tickets were purchased or through UCR. How much would you pay for a policy which pays $100,000 in case of your death due to terrorism / any cause?
• Your data, $100k policy for death due to: – Terrorism: – Any cause:
More on Unpacking
1. What is the probability that it will rain in Riverside during finals week this quarter?
2. a. What is the probability that it will rain in Riverside exactly one day during finals week this quarter? b. What is the probability that it will rain in Riverside more than one day during finals week this quarter?
Event Average Probability
Packed 36%
Unpacked ‐ One day ‐ Two or more days
75% 50% 25%
BUS143 Topic 2
10
Some marketing implications
To make things appear more likely or larger:
• Create familiarity, especially right before consumers make decisions – This is why Google search ads are so powerful – And why Facebook ads are effective
• Link to pre-existing knowledge structures (the power of stories)
• Use vivid imagery
– Encourage customers to mentally imagine the experience
• How might store music affect your purchasing behavior?
Subtle uses of availability: Priming
0
5
10
15
20
25
30
35
40
45
Buy French Wine Buy German Wine
Hear French Music
Hear German Music
BUS143 Topic 2
11
Unintentional use of availability: Mere Measurement (Morwitz, Johnson, & Schmittlein, 1993)
0
1
2
3
4
5
Computers Cars
% B u y in g P ro d u ct
A 33% increase
An 18% increase
Availability Wrap-Up
• Familiarity, recency, and vividness (or the lack thereof) affect judgments and behavior – Overestimate salient causes of death, earnings of
familiar companies – Used extensively in advertising
• Availability can impact choice without awareness – Priming: German vs. French wine
BUS143 Topic 2
12
Representativeness: a.k.a. “the Similarity Heuristic”
In forming a judgment, we tend to… • Make predictions and evaluations based on
similarity to salient examples and schemas (i.e, mental frameworks of the world)
• More simply: People draw analogies to what they think is a similar situation or a good example
Problems with representativeness
• Problem 1: It’s extremely sensitive to the example(s) selected – People are anecdote rather than data-
driven – Favors recent and vivid examples
(i.e., availability), and stereotypes – Many bad examples! Atypical (unlikely to happen again)
E.g., UVA losing in the first round of NCAA tournament
Only superficial similarity E.g, product name and packaging color
BUS143 Topic 2
13
Problems with representativeness
• Problem 2: Beliefs about causes of random outcomes are often not quite right – Base-rate neglect (remember: Flu problem) – Traditional medicine (e.g., Chinese) uses for rhino
horn, tiger penis, bear testicles, etc. – Misunderstanding how randomness works “Regression to the mean” Seeing randomness where it’s not, not seeing randomness
where it is
Forecasting Problem (in MobLab) Cox & Summers 1987
Enter as millions (no need for the 0’s), make sure it adds to 99!
BUS143 Topic 2
14
Even experts forget to be regressive
22 of 35 “excellent” companies underperformed the S&P 500 over next 10 years (e.g., Atari, Wang Labs) Maybe not so great after all…
Only 8 of 18 “visionary” companies outperformed the S&P 500 over next 10 years
Representativeness: Misperceptions of randomness
P(switch) = .37 P(switch) = .51 P(switch) = .63
Streaks don’t feel representative of randomness! (think streaks of same answers on a multiple choice test)
BUS143 Topic 2
15
Representativeness: Streaks and gambler’s fallacy
• Suppose you’re playing roulette. The ball landed on red 5 times in a row. What do you want to bet on?
A. Bet it all on red! B. Bet it all on black!
• This is the same principle that makes music playlists not feel random enough on “shuffle”! • iPod, Spotify users complained playlists don’t seem random
New algorithm that spaces out artists more evenly
Real World Implications
• Choose examples and analogies wisely…
• Shape people’s evaluations by influencing associations – Increase availability of beneficial examples – Increase genuine or superficial similarity to certain
examples
• Don’t trust your intuitions about randomness – Remember about regression to the mean!
BUS143 Topic 2
16
Aside: Why do heuristics persist?
• Big reason: Confirmation bias • Types of confirmation bias – Selective Search: Seeking information that confirms
(both purposely and non-purposely) hypothesis Talk to friends that you expect to agree with you, asking
leading questions (Don’t you love BUS143?)
– Interpret ambiguous info in line with hypothesis E.g., horoscopes, fortune tellers
– Biased memory E.g., remembering a robber as being taller, bigger than in
reality
Wason’s Card Task
Suppose each card has a number on one side and a letter on the other. Which of these card(s) are worth turning over if you want to know whether the statement below is false? "If a card has a vowel on one side, then it has an even number on the other side."
BUS143 Topic 2
17
Wason’s Card Task – Now in context
Imagine you’re a bouncer at a bar. You must enforce the rule that if a person is drinking beer, then he or she must be over 21 years old. The four cards below each represent one customer in your bar. One side shows what the person is drinking, and the other side shows the drinker’s age. Pick only the cards you definitely need to turn over to see if any of these people are breaking the law and need to be thrown out.
Can a smell help avoid confirmation bias?
• Lee & Schwarz (2012) found that exposing people to incidental fishy smells made them more suspicious
More likely to try disconfirming triplets (the 4, 8, 12 game)!
• Example of embodied cognition
• Another ‘intuitive’ way to be more disconfirming? Treat everyday like April Fool’s Day!
BUS143 Topic 2
18
Disconfirmation Practice: CEOs and their Pets
“Results of a recent survey of 74 chief executive officers indicate that there may be a link between childhood pet ownership and future career success.
Fully 94% of the CEOs, all of them employed within Fortune 500 companies, had possessed a dog, a cat, or both as youngsters….
The respondents said that pet ownership helped them develop many of the positive character traits that make them good managers today, including responsibility, empathy, generosity, and good communication skills.”
Management Focus Magazine
What do you think? Are pets important for CEOs?
Anchoring and Adjustment
In forming a judgment, we tend to… • Use starting values (“anchors”) and adjust our
judgment in the direction that seems appropriate
• Anchoring works by ‘unconsciously’ increasing the availability of some information
• Many problems: – People are not aware of anchors – People use even irrelevant anchors – People do not adjust enough from the anchor – (Like other heuristics) Can lead to overconfidence
BUS143 Topic 2
19
Anchoring Example: Provided anchor
1. What is the probability that 2019 Toyota Camry passenger car sales (in the United States) were higher than 100,000 (1,000,000)?
2. What is your best guess (in thousands of cars) as to 2019 Toyota Camry passenger car sales (in the United States)?
• 100,000 anchor: 200,000 • 1,000,000 anchor: 750,000 • Actual: 336,978
Bonus: Best selling cars of 2019 (USA) 1. Ford F-Series 896,526 2. Dodge Ram 633,694 3. Chevrolet Silverado 575,600 4. Toyota Rav 4 448,071 5. Honda CR-V 384,168 6. Nissan Rogue 350,447 7. Chevrolet Equinox 346,048
Anchoring Example: Unit anchors
• Estimate the total U.S. egg production in 2019. – in billions – in millions
• Billions: 20 billion • Millions: 300 million • Actual: 95.3 billion
Bonus fact: Average American eats ~280 eggs a year!
BUS143 Topic 2
20
Anchoring Example: Even totally uninformative anchors
• What are the last three digits of your cell phone number? 446 • Would you pay that much for an iPad Pro 64gb? 70% said yes • What is the most you would pay?
r = .39
$407 vs. $600 (p < .001)
Real World Examples?
• Pricing – Sales prices
• Suggested quantities
• Predictions of tastes – “False consensus” effect
BUS143 Topic 2
21
Do credit card minimum payments anchor? Stewart, 2009
• Minimum Payment
• For people making a partial payment, r = .75 correlation between minimum payment and actual payment amount
• If minimum payment is removed, payments rose by 70%!
Experts are NOT immune, and the consequences can be huge
Listed Price (Anchor)
Estimates by Real Estate Agents
Appraised value
Recommended Selling Price
Reasonable Purchase Price
$129,900 $114,204 $117,745 $111,454
$139,900 $125,041 $128,530 $124,653
$149,000 $128,754 $130,981 $127,318
BUS143 Topic 2
22
Wrap-up of Heuristics
• Availability, representativeness, and anchoring over-weighing information
• Quality of the information (sample size; validity) is under-weighed
Things to Remember
• Effective marketing (persuasion, PR) means getting your ideas in people’s heads…
• And on careful selection of those ideas… – Even superficial similarity to examples can
powerfully influence liking
– Even somewhat arbitrarily suggested numbers (asking prices, suggested quantities, yesterday’s trading value) influence prediction, valuation, and choice
BUS143 Topic 2
23
Ye’s Keys
4. Recent, vivid, and/or familiar examples are easy recalled and this feeling of availability impacts judgments, often without awareness.
5. People draw analogies to representative examples and fit data to patterns, leading to biased judgments, especially of randomness.
6. Numbers—even completely irrelevant ones—can anchor numerical judgments.
7. Confirmation bias—the tendency to focus on information consistent with a favored hypothesis and ignore information consistent with other hypotheses—makes these biases hard to avoid.
