You are on page 1of 35

INFORMATION IS DESIRED, EVEN IF USELESS

ELLSBERG, 1961

 There are two opaque urns


 Urn I contains 50 red balls and 50 black balls
 Urn 1I contains 100 balls, either red or black, with unknown numbers of each type.The total number
of balls in this urn is 100.There are no other balls (i.e. not red or black) inside.

 You task is to:


 1) Select the urn you want to draw from (Urn I or II)
 2) Select the color of the ball you think will be drawn (Red or Black)

 A ball will then be drawn from the urn of your choice. If its color matches your choice, you get $100
THE ELLSBERG PARADOX

 Participants show the following preferences:


 Indifference between choosing red or black balls from Urn 1
 Indifference between choosing red or black balls from Urn II
 Strong preference to draw from Urn I

 Why i s that problematic?


THE ELLSBERG PARADOX

 We know that Urn I has a 50/50 chance of drawing a red or black ball
 In Urn II,
 If participants suspect there to be more black than red balls, they should prefer Urn II and Black
color
 If participants suspect there to be more red than black balls, they should prefer Urn II and Red
color
 If participants suspect Urn II to be 50/50, they should be indifferent between Urn I and Urn II

 …but they prefer Urn I


 The indifference between Red and Black suggests they estimate the chances for Urn II to be 50/50
THE ELLSBERG PARADOX

 Moreover Urn II is probabilistically not different from Urn I. Since we do


not know the distribution of red and black balls, the domain of all
possible distributions (0R,100B - 100R,0B) produces a 50/50 chance of
guessing
AMBIGUITY AVERSION

 The Ellsberg paradox demonstrates one thing:


We don’t like the unknown
 This is called ambiguity aversion: a preference for known (even if risky)
probabilities over unknown probabilities
 The knowledge of probabilities (even as un-guaranteed as 50/50) makes
us feel safer
REASONS FOR AMBIGUITY AVERSION

 Segal (1987) suggested that a lottery with unknown probabilities is actually a two-
stage lottery for the participant
 Stage 1 is imaginary:The decision maker considers the urn as one of the set of urns
(the urn is ‚drawn’ from the possible distribution). For example the participant may
guess that the Ellsberg ambiguous urn contains 30 red and 70 black balls.
 Stage 2 consists in playing the lottery based on the assumption from Stage I (the
participant would choose black, as they assume this is the distribution)
 However, since the participant has no real argument as to why the distribution
should be 30/70, there is additional uncertainty from the other possible scenarios (I
bet on black and what if all balls are red?)
REASONS FOR AMBIGUITY AVERSION (AL-NAJJAR & WEINSTEIN,
2009)

 The fear of being cheated: Agents may assume that the ambiguous probability is
adversely manipulable (i.e. the distribution within the urn will change after I state my
bet). In this case the experimenter is an opponent in a zero-sum game. It is therefore
safer to bet on known probabilities than to allow the possibility that Urn II will be
manipulated to our detriment
 Myerson, 1991: „People offer bets only when they have some special information [which
gives them an edge]”
 Misapplied heuristics: the inability to apply probability theory instead of heuristics is
correlated with ambiguity aversion (Halevy, 2007)
 We mistake the lottery for a competition game, assume the other party has more information than
us, possibly based on a misapplied representativeness heuristic
THE SURE-THING PRINCIPLE (SAVAGE, 1954)

 There are choices where the agent:


 Prefers outcome 1 over 2 knowing that X happened/will happen
 Prefers outcome 1 over 2 knowing that X did not/will not happen

 In such a case, the agent should prefer 1 over 2 not knowing whether X happened
 Imagine you play a coin-toss where you get $200 if heads turn up and
lose $100 when tails turn up.
 Assume you just played and won $200. Would you play again?
 Imagine you play a coin-toss where you get $200 if heads turn up and
lose $100 when tails turn up.
 Assume you just played and lost $100. Would you play again?
 Imagine you play a coin-toss where you get $200 if heads turn up and
lose $100 when tails turn up.
 Assume you just played, but do not yet know the result of the gamble. Would you
play again?
THE DISJUNCTION EFFECT (TVERSKY & SHAFIR, 1992)

 In their research:
 69% of subjects who WON wanted to play again
 59% of subjects who LOST wanted to play again
 Only 36% of subjects who DID NOT KNOW wanted to play again

 So the sure-thing principle is violated


THE DISJUNCTION EFFECT

 Tversky & Shafir explain it as reason-based analysis – the reasons for playing again are as
follows:
 When you won, you play for house money and a loss in the second gamble still leaves you in the domain
of gains
 When you lost, you play to break even and a win in the second game brings you back to the domaing of
gains

 Without knowledge of the outcome of the first gamble, neither of the above reasons apply.There is no
clear motivation to seek risk.
 Moreover, there is more uncertainty as to the outcome: the agent’s expected utility of gambling in the
second lottery (additive) varies from -200 to +400 for UNKNOWN, while it is +100/+400 for WON
and -100/+100 for LOST
BAGASSI & MACCHI (2006): INFORMATION-DEPENDENT APPROACH

 In the original research by Tversky & Shafir, the statement (UNKNOWN) was
you will not know whether you have won $200 or lost $100 until you make a
decision…
 This could implicate playing to know rather than playing to win
 When the statement was modified to you do not know whether you have won
$200 or lost $100.You are now offered a second identical gamble, the effect
disappeared! (42% willing to gamble in the original, 60% willing to gamble in
the modified version)
 The aim of the second gamble was moved (in the original) from winning to knowing, and
not gambling is the safer way of finding out.
AND NOW FOR SOMETHING COMPLETELY DIFFERENT

 When betting on a national lottery, would you prefer to:


 A) let the numbers be randomly generated, or
 B) select you own numbers?
 When playing roulette, would you
prefer
 A) spinning the wheel yourself, or
 B) letting the croupier spin the
wheel?
 Goffman (1967) observed that casino croupiers (dealers) experiencing bad luck run
the risk of losing their jobs
 Henslin (1967) observed that dice players try to control the outcome by throwing
the dice softly for low numbers and throwing them hard for high numbers
 Strickler, Lewicki & Katz (1966): Subjects tend to bet more in a game of dice if the
bet is made before the toss than if the bet is made after the toss (but before the
outcome is known)

 This phenomenon is known as the illusion of control – overestimating one’s influence


on events decided by random chance
ILLUSION OF CONTROL (LANGER, 1975)

 Thinking of random chance events as skill-dependent


 Using luck and ‚luck-inducing’ rituals to achieve the desired outcome
 Unrealistically optimistic predictions of success (Langer & Roth, 1975;
McKenna, 1993)
FACTORS ACTIVATING ILLUSION OF CONTROL

 The task should be presented as skill-based and controllable


 Choice
 Stimulus/Response familiarity
 Active involvement
 Competition
ILLUSION OF CONTROL: COMPETITION

 Competition creates illusion of control as


a function of the other participant’s skill
compared to our own (or beliefs thereof)
 Competing against an opponent with low skill,
illusion of control is achieved
 Participants competing against a nervous,
awkward confederate bet more (nearly twice
as much) than when competing against a
confident one
 …in a game which consisted in drawing the highest
card from a deck.
ILLUSION OF CONTROL: CHOICE

 Subjects participated in a lottery, where a ticket would be drawn from an urn


by a third party. Participants either:
 Could choose the ticket for themselves, or
 Were handed a ticket by another participant
 Afterwards subjects were asked whther they want to sell their tickets, and for
how much (the initial price of the ticket was $1)
 In the choice condition, the mean price was $8.67
 In the no-choice condition it was $1.96
ILLUSION OF CONTROL: FAMILIARITY

 Following the last experiment, participants were


asked whether they wanted to swap the lottery
ticket for a better one (with much better
chances of winning), where the new ticket
contained either
 A) one of 26 letters of the alphabet
 B) one of 26 unfamiliar esoteric symbols

 Participants were more likely to switch from the


original lottery to the letters of the alphabet, and less
likely to swap for the unfamiliar symbols
ILLUSION OF CONTROL: PASSIVE INVOLVEMENT

 Participants were given lottery tickets at the beginning of a horse track day.
The lottery was regularly reminded to the participants using radio
announcements.
 The longer the participants waited for the lottery (i.e. the more time passed since they
received the ticket), the more confident they were about winning
 In another study, subjects were either given a three-digit lottery ticket at once
or given three parts with one number each in installments over time. The
latter condition (more involvement) led to a higher belief in winning
ILLUSION OF CONTROL BY PROXY (WOHL & ENZLE, 2009)

 Agents also use other people as means to achieve illusory control


 A lottery ticket was to be chosen either by the participant or by another person
 The other person was either presented as lucky or neutral.
 Agents were more likely to ask lucky confederates to choose for them, and
choose themselves when paired with neutral confederates
WHY THIS ALL HAPPENS: CONFIRMATION BIAS AND ILLUSION OF
CORRELATION

 The human mind hates the unknown and strives to explain everything it
encounters. Randomness is not an acceptable answer
 Confirmation bias makes us search for data wchich confirms our prior
assumptions, rather than data which falsifies it. This leads to a sense of
control and reduces perceived ambiguity.
CONFIRMATION BIAS (PLOUS, 1993)

 When testing a belief:


 We search for information which confirms it
 We ignore information which contradicts it

 We interpret ambiguous information in a way that confirms our prior belief

 Nickerson (1998) calls confirmation bias a spontaneously built one-sided case, i.e. the selectivity of
information processing is not deliberate, but unwitting (automatic)
 We are motivated to confirm our prior beliefs in order to avoid congitive dissonance, easily achieve
closure, retain our self-image (self-verification), seem rational in the eyes of ourselves and the others,
avoid regret, reduce the feeling of loss to incorrectly made decisions, reduce ambiguity, etc.
CONFIRMATION BIAS

 Bruner, Goodnow & Austin (1956): participants test a hypothesis by looking for
positive cases (instances of the hypothesis being true)
 I saw 10 men drinking beer. Therefore all men drink beer.
 Levine (1970): participants search for members of the tested
categories/populations, ignoring non-members as reference (i.e. ignoring
elements inconsistent with the hypothesis)
 Wason task (1966): Test the rule that if a card has an even number on one side, then there
is beer on the opposite side?
 You are allowed to turn 2 cards
CONFIRMATION BIAS

 Overweighting the importance of positive


instances, underweighting the negative
ones
 Pyszczynski & Greenberg (1987) – subjects
need less positive evidence to accept a
hypothesis than negative evidence to reject it
 …if you believe that mentalism is real, you will
find support in what a mind-reader says that
shows their ability to read minds, even if they
cold-guess and are 50% accurate
 Same goes for fortune tellers
CONFIRMATION BIAS IN LOTTERIES

 An urn contains one of two proportions of red, green and black balls. You
presume there are 40 black, 40 red and 20 green balls. The alternative is
that there are 40 black, 20 red and 40 green balls. How would you react
to a black ball being drawn by another participant? Does it confirm your
hypothesis? Are you more certain of the distribution now?

 Pitz, 1969; Troutman & Shanteau, 1977


ILLUSORY CORRELATION AND ILLUSORY PATTERNS

 Determined to find patterns and explain the world, we tend to detect


correlations which do not exist (Mullen & Johnson, 1990)
 Plous, 1993 demonstrated that we tend to overemphasize double positives:
 A black cat crossed your path, and then you lost your wallet. (this is a double
positive). Therefore you have proof that black cats bring bad luck
 …but how many times were you unlucky when a cat did not cross your path?
How many times did a cat cross your path and you were not unlucky? These
elements are ignored.
ILLUSORY CORRELATION

 Searching for patterns, supported by the confirmation bias leads to the


following:
 Superstitions
 Illusory control
 Beliefs in good and bad luck
 Belief in karma
 Private theories on reducing ambiguity in objectively ambiguous situations

 Illusory correlation is a manifestation of ambiguity aversion


ILLUSORY CORRELATION IN GAMBLING

 Delfabbro, Lahn & Grabosky (2006) investigated


causes of pathological gambling in adolescents
 Problem gamblers overestimate the role of skill, and the
perceived chance to turn a profit in gambling
 Illusion of control in gamblers stems from illusory
correlations
 When faced with important tasks, we tend to
ignore objective data in favor of our subjective
experience
 Therefore trying to replicate actions and context of prior
instances where we won
 And avoiding replication of contexts when we lost
ILLUSION OF VALIDITY

 Are traders able to pick stocks which will generate profits?


 Kahneman (2011): illusory skill; symmetrical market with identical information for
buyers and sellers
 Correlation between returns year to year is .01

 Odean (1998): unrealistic optimism in traders


 How much you expect to earn vs How much you earn
 Overestimation by 0.5-1% per 2-4 weeks, and by 3.2% per year (not percentage of earnings –
percentage of investment!)
ILLUSION OF VALIDITY

 Treating luck as skill; rewards, bonuses & job loss dependent on luck
 Can one be „the best” at roulette? Or „the best” at coin toss?
 Books on „How to gamble” and „How to beat the stockmarket” are sold in millions of copies

 Reason for illusory skill: work investment and exercising high-level skills (even if they are uncorrelated with the outcome)
 „Gambling technique”
 „Day trading strategy”
 „Economic knowledge”
 „Advanced prediction methods”
 …magic, clairvoyance, fortune telling, horoscopes, tarot – they all seem like the individual has high knowledge and abilities
 Illusory correlation between skill and lucky outcome

We search for predictability even where there is none

You might also like