Decision-making, heuristics and biases

1.

Decision-making, heuristics
and biases
[email protected]
@drdphd

2.

Three layers
• Behavior (including mimics and speech = timbre + text)
• Cognitive/Affective
• Neural
• Well-known
• Valence
• Arousal
• Cognitive load
• Proprietary or not-stabilized in the theory yeet
• Suspense
• Immersion
• Being convinced
• …

3.

4.

5.

Simple classification

6.

How people are making decisions
Neoclassical “benchmark”
H.Simon line
• Selfishness, lack of emotions
• Doesn't regret, doesn't change his mind
• Choosing the best option
• Comparison of costs/benefits of all
possible options
• Has clear and stable preferences that are
not influenced by others
• Knows the results of all possible options
• Updates beliefs based on new
information
• Experience = “deep learning”
• Bounded rationality
• Search is not free, cognitive
resources are limited
• Heuristics
• First: satisficing (Simon, 1957)
• Universal "rules of thumb"
• Saves cognitive resources and is
almost always suitable

7.

Dual systems theory
«System 1»
«System 2»
• Intuitively
• Fast
• Automatically
• Easily
• The process is hidden
• Sensitivity to emotions
• Slowly
• Consciously
• With intellectual effort
• The process is “spoken”
• Logical

8.

Where neoclassical man misses
• Number of selection options
• Self-knowledge = learning about own preferences
• Self-control and preference stability

9.

10.

11.

12.

13.

How beliefs are made

14.

Heuristics
• decision-making rules are developed through trial and error,
experiments, personal experience
• decisions are rarely made based on a mental model of some process
• Shefrin (2011):
• Using heuristics (i.e., rules of thumb) to draw conclusions from the
information available
• People are sometimes prone to certain mistakes because
• their heuristics are imperfect
• decision-making situations have a certain format

15.

Definitions
• Heuristics – techniques for simplifying the analysis of complex
situations and probabilities
• Unlike an algorithm, a heuristic does not rely on a model of a phenomenon,
but routes a mental process or creates a rule to solve a problem by
simplifying information
• Deviation (bias) is a predisposition of consciousness to certain stable
processes leading to decisions other than “rational” ones
• Many deviations are a consequence of heuristics
• Effect is a synonym for deviation

16.

What table is longer?
Awareness of biases does not eliminate exposure to them

17.

Availability heuristics

18.

Why people die (rank)
• Nutrition problems
• Lung diseases (infections, oncology)
• War and civil conflicts

19.

What are the most words in English?
• Starting with "A"
• Having a third letter "A"

20.

What are the most words in Russian?
• Abstract (“love”, “thought”)
• Specific (“door”, “water”)

21.

22.

Which pairs of words were found more often?

23.

Availability heuristic
• The tendency to rely on knowledge that is readily available rather
than exploring other alternatives or procedures.
• For example, experience
• our memory is influenced by information unrelated to the objective
frequency of actual events (actual base rates).
• Elements of large classes are recalled better and faster
• More likely events are easier to imagine
• Associative connections are strengthened when events occur simultaneously

24.

Availability heuristic
• Logical procedure, but not always all memories are “available”,
according to Kahneman, Tversky (1974)
• Priority is given to recent, memorable/vivid information
• When assessing the likelihood of an event—crash in a plane crash—
people often use stories they know personally.
• “Personal” events are better remembered - the death of loved ones in
a plane crash, for example
• Conclusions supported by a larger number of known facts seem more
reliable (Mukhortov)

25.

26.

Biases due to availability heuristics
• Forming an opinion based on
• available facts (plane crashes)
• ineffective samples (letter “r”)
• imaginary probabilities (taking into account all risks without taking into
account their probabilities)
• illusory correlations

27.

Causes of death
Rank
Cause
Dead in 2015
3
Война и гражданские конфликты
182 000
2
Проблемы с питанием
418 000
1
Болезни легких (инфекции, онкология)
3 500 000

28.

Event recoverability
• AKA Memory Brightness
• Estimation of causes of mortality
• Fear of airplanes
• Airline Security Solutions
• The only “shoe bomb” -> everyone takes off their shoes
• Performance ratings
• managers evaluating employee performance depend on performance
over the past 6 months

29.

Search direction efficiency
• Easy to retrieve from memory
• Words starting with "a" and words with "a" as the third letter
• Hiring decisions
• Managers are more likely to hire people they know and pass over
more qualified candidates

30.

Ability to imagine images
• The risk of a dangerous expedition is assessed through mental
reproduction of unforeseen circumstances
• If the danger is difficult to imagine, the risk will be underestimated

31.

Representativeness heuristics

32.

Experiment 1
Anna. 35 years.
Intelligent, unmarried, graduated with honors from
the Faculty of Philosophy.
During her student years she participated in political
demonstrations.
Which is more plausible:
Anna works as a teller in a bank.
Anna is a feminist and works as a bank teller.

33.

Experiment
1000 people match the description:
Anna. 35 years.
Intelligent, unmarried, graduated with honors
from the Faculty of Philosophy.
During her student years she participated in
political demonstrations.
Which is more plausible:
Anna works as a teller in a bank.
Anna is a feminist and works as a bank teller.

34.

Experiment
Lisa is 31 and pregnant. She is concerned about birth defects such as
Down syndrome. Her doctor tells her that she doesn't need to worry
too much because there is only a 1 in 1,000 chance that a woman
her age will have a child with Down syndrome.
When a child has Down syndrome, the test is positive 86% of the
time. There is also a small "false positive" rate: 5% of babies test
positive despite not having Down syndrome. Lisa takes the test and
gets a positive result. Given this test result, what are the chances
that her baby has Down syndrome?

35.

Some city is served by two hospitals - a large one and
a small one. At the larger hospital, approximately 45
babies are born each day, and at the smaller hospital,
approximately 15 babies are born.
As you know, approximately 50% of all babies are
boys. The exact percentage of babies who are boys,
however, varies every day. Sometimes it can be
higher than 50%, sometimes lower.
Over the course of 1 year, each hospital recorded the
days on which (more/less than) 60% of the infants
born were male. Which hospital do you think has
recorded more days like this?

36.

Imagine a basket filled with balls, 2/3 of which are one
color and 1/3 of another.
One person takes 5 balls from a basket and discovers
that 4 of them are red and 1 is white.
Another person takes out 20 balls and discovers that 12
of them are red and 8 are white.
Which of these two people should be more confident in
saying that the basket contains 2/3 red balls and 1/3
white balls rather than vice versa?
What are the chances for each of these people?

37.

Representativeness heuristic
• Stereotyping: object A2 belongs to sample A because it is similar to
A1
• intuition about sample results
• typical for both professionals and amateurs
• often fails due to fundamental problems
• arbitrary analogies in situations where a decision is made about a
little-known/obscure phenomenon, and there is a “similar” wellknown model

38.

39.

40.

Base rate neglect
• People use Bayes' law incorrectly, exaggerating the probability
(description | FinB) and ignoring the probability (BinB) – base rate

41.

Representativeness and Linda
• P(A & B) ≤ P(A)
• Gigerenzer, 1993, 91% = B, 22% - B
• Base rate neglect

42.

Liza
• What are the chances that her baby has Down syndrome? 1.7%
• For every 1,000 women Lisa's age, 999 do not have a child with Down syndrome.
• 5% of women with a child who do not have Down syndrome will receive a false
positive test.
• 999 * .05 = 49.95 women per 1000 people get a false result.
• 86% of children with Down syndrome will test positive the first time.
• .86 * 1 = .86 per 1000 women with a Down syndrome child receive an accurate
positive test
• .86 / (86 + 49.95) = 1.7% of women who test positive have a child with Down
syndrome
• People don't consider the fact that only 1/1000 women Linda's age get an
accurate false result

43.

Sample Size bias
• Part of representativenes
• When assessing the likelihood that data was generated by a particular
model, people do not take into account the sample size - it may be
small
• 6 rolls 3/3 = 1000 information content
• Making quick (and wrong) decisions about the process

44.

Gamblers fallacy
• “Dostoevsky's system”
• “Errors balance each other out in a self-correcting process.”
• This is true, but does not apply to limited samples
• In large samples this is known as the "law of large numbers"
• Contrarian investing

45.

Big numbers law
Source: Aronson

46.

Small numbers law
• The belief that even small samples reflect the characteristics of the
population is sometimes called the “law of small numbers” [Rabin
(2002)]
• A financial analyst with 4 good recommendations is “not like” a bad
financial analyst
• The phenomenon of “hot hands”: fans believe that a basketball player
who has made 3 successful shots will soon make another successful
one, even if this is not confirmed by other data [Gilovich, Vallone,
Tversky (1985)]

47.

Conservatism heuristics

48.

https://bit.ly/ranepaNeuro

49.

Conservatism heuristics
• Conservatism: once people form a belief, they become attached to it seriously and for a
long time [Lord, Ross, Lepper (1979)]
• Judgments are made by confirming previous information rather than seeking new
evidence
• The opposite of base rate neglect
• If data “looks like” a particular case (model), people overestimate the significance of the data,
underestimating the more general model
• If the data doesn't look like anything, people completely ignore the data and rely only on the
general model
• belief perseverance: refusal to update one's belief if the flaws in it are already known
• three effects:
• avoidance of dissonance (Festinger)
• skepticism if it cannot be avoided
• confirmation bias – reverse interpretation of disproving information
• Festinger, the story of the sect and the end of the world

50.

Beliefs perseverance
“Faced with the choice between
changing one’s mind and proving
that there is no need to do so,
almost everyone gets busy on the
proof”

51.

Biases
• Acknowledgment Rejection
• Anchoring
• Conjunctive and disjunctive events
• Overconfidence
• "Hindsight“
• Disposition

52.

Anchoring
• People proceed from initial approaches and assessments, becoming
emotionally attached to them
• People love their thoughts, the comfort of the known
• Anchoring is the inability to fully take into account (adjust to) the
influence of new information (i.e., a manifestation of conservatism)
• Car dealers set a high starting price and then begin to lower the price
from there (Schiller)
• Hammer Man Syndrome (Munger)

53.

Anchoring
• Lack of adaptability – failure to adjust the basis for comparison
• Reassessment of conjunctive probabilities – simultaneous “triggering”
of associated risks
• Underestimation of disjunctive probabilities
• “wishful thinking” = “wishful thinking”

54.

Overconfidence
• excessive confidence, overconfidence
• Most people have unrealistically rosy beliefs about their abilities and
prospects (Weinstein (1980))
• More than 90% of respondents think they are above average in areas such
as driving ability, sense of humor and ability to get along with people.
• They also demonstrate systematic errors in planning, predicting that tasks
(such as writing essays) will be completed much earlier than they actually
are (Buehler, Griffin, Ross (1994)).
• The less complex the task, the less self-confident the subjects are (Vine)
• The better the subjects knew the subject, the less self-confident they
became (Vine)

55.

Context-dependence

56.

Experiment
A ticket to a concert of your favorite band costs $100. On the subway on the
way to a concert, you discover that your ticket has been stolen. I have $100 and
a travel card in my pocket.
You can buy a new ticket before entering the concert. Will you buy it?
1. Yes
2. No

57.

Experiment
A ticket to a concert of your favorite band costs $100. On the subway on the
way to a concert, you discover that your ticket has been stolen. I have $100
and a travel card in my pocket.
You can buy a new ticket before entering the concert. Will you buy it?
1. Yes
2. No

58.

Mental accounting
• Ticket experiment
• The most important premise of prospect theory
• Shows that people perceive losses and gains differently
• This means that depending on how we describe the problem, the
solution may be different.
• Narrow framing - the tendency to separate various games from the
whole wealth
• You play the game 5 times but don't know the outcome; will you play
for the 6th time? 60% of individuals refused.

59.

Loss aversion
• Refusal to record a paper loss
• Avoidance of losses can lead to risk-seeking behavior (risk propensity)
• The trader made a number of unprofitable trades in the first half of
the quarter
• Will try to make up for it in the second half

60.

Loss aversion – scarcity effect
• Subjects judge cookies from a jar that contains less of them to be
tastier, even though the cookies are the same (Worchel, Lee, and
Eidwall, 1975);
• In dynamics, the value of a product falls if it is no longer scarce;
• How to use it: on the product page write “Only 3 pieces left in stock.”

61.

62.

Prospect theory

63.

Elements of the theory
• Entities: games, lotteries, winnings, wealth, prospects
• Decision making is carried out in two stages - editing / framing and
evaluation phase
• Editing a selection situation is carried out using “standard” operations
• Game choice (perspective assessment) is based on special functions
for weighing probabilities and outcomes
• Preference discounting is a later addition: time and learning emerge.

64.

Framing
• Because of framing, the solution depends on how the problem is
described.
• Rational choice assumes that the decision, regardless of the
description
• When describing a problem, a person continues the story with
heuristics
• Depending on the order of the phases, outcomes may vary

65.

Editing operations
• Coding: selecting a reference point
• Combination: (200, .25; 200, .25) ⇒ (200, .5)
• Segregation – separation of the risk-free component: (300, .8; 200, .2)
⇒ (200) + (100, .8)
• Cancellation: Ignoring common aspects of games when comparing
them to each other
• Simplification: rounding/ignoring
• Determining the dominant: highlighting the most preferred game

66.

Game utility
• Participation in the lottery win/lose is assessed not in terms of
absolute values, but in terms of win/loss
• The utility of the game is defined as
• Subjective probability of winning * subjective winning size
• MINUS
• Subjective probability of loss * subjective size of loss

67.

68.

69.

Preference discounting

70.

Preference discounting

71.

Preference discounting