“deep” belief

advertisement
Self-deception as self-signaling:
a model and experimental evidence
Kyle Krueger, Michael Ritchie, and Lance Braud
Self-Deception

Gur & Sackeim’s (1979) definition
An individual:
 holds
two contradictory beliefs
 holds them simultaneously
 is unaware of holding one of the beliefs
 is motivated to remain unaware of that belief
Motivational Bias Explanation


There is no conscious intention to self-deceive.
Instead, an individual makes judgments based on
unconscious motivations.
Examples
 Self-serving
bias
 Attributing
successes to internal factors, and failures to
external factors
 Confirmation
 Tendency
bias
to interpret events (our own actions) in ways that
confirm our beliefs
Self-Signaling Theory



Our choices affect our beliefs
A choice not only causes an action, it also expresses
a belief
Levels of Belief
 Deep
belief
 State belief
 Experienced belief
Self-Signaling Theory

Levels of Belief
 Deep
belief
 Stated belief
 Experienced belief


If stated belief does not match deep belief, then
there is attempted self-deception
If experienced belief is equal to stated belief, it is
a successful self-deception.
Self-Signaling vs Motivational Bias

Motivational bias does not account for three
characteristics of self-deception
 Defensiveness
associated with challenged beliefs
 The special significance of beliefs about the self
 You
wouldn’t state that someone was guilty of self-deception
if they were convinced the moon landing wasn’t real
 Self-deception
A
can fail
bias should necessarily alter belief
The Self-Signaling Formal Model

Bodner & Prelec (1995)
 Model

of non-causal motivation
Three primary assumptions
1. There exists a characteristic that is personally
important
2. This is not possible to introspectively evaluate
3. It can be examined through one’s actions
The Formal Model of Self-Signaling
V(x, θO) = u(x, θO) + λ∑θu(x,θ)p(θ|x)
The Self-Signaling Formal Model
V(x, θO) =
u(x, θO) +
λ∑θu(x,θ)p(θ|x)
The total utility of an
action is equal to
the generated utility
of an action, plus
the diagnostic utility of
an action




x = the action or outcome
θO = the “deep” belief about some characteristic
θ = the experienced belief about some characteristic
λ = the individual’s level of self-deception
Diagnostic Utility
λ∑θu(x,θ)p(θ|x)

This must account for two emerging paradoxes
1. The static state paradox
The ability to hold two opposing beliefs
 Explained by our two separate terms, θO and θ

2. The dynamic paradox
The ability to remain unaware of a held belief
 To explain this, there must be two variants of self-signaling



Face-value
Rational
Diagnostic Utility
λ∑θu(x,θ)p(θ|x)

Face-value
Our knowledge of θO is a probability distribution, p(θ)
 p(θ|x) = our inferred value of θO given the action x


This is derived with the assumption that our action is to solely
provide maximum outcome utility

“By choosing x I demonstrate deep beliefs such that x maximizes
standard expected utility given these deep beliefs”
Diagnostic Utility
λ∑θu(x,θ)p(θ|x)

Rational


Inferences about motivation are exactly the same as in an
interpersonal scenario
Only paradoxical if action and belief are inconsistent


As such, belief θ influences the action x, and action x derives belief θ
Not ignorant of diagnostic utility and so can be mitigated
rationally. This variation is through λ
The Formal Model Revisited
V(x, θO) =
u(x, θO) +
λ∑θu(x,θ)p(θ|x)
The total utility of the outcome utility The diagnostic utility of an
action, as determined by
an action
given
the
of
an
action
susceptibility of selfx = the action or outcome
deception, the utility of an
deep belief
of
a
given
a
deep
 θO = the “deep” belief about some action
characteristic
given an experienced
characteristic is
belief, plus
belief, and the change in our
equal to
experienced belief given
that action
The Experiment
Purpose
 Designed to induce self deception in
individuals, where self deception has a cost.
 Designed to examine failed self deception,
where the subject is aware of their own
attempted self deception.
Design
Female rated symbols
Male rated symbols
Phase one
85 subjects view 100 Korean symbols.
 The subject classifies the symbol as male or
female and then rates their confidence on a 5
point scale.
 Subjects are awarded $0.02 for every correct
classification.

Phase Two
The subjects are asked to predict the gender
of a symbol, then they are shown the symbol
and asked to confirm or reject their prediction.
 Again subjects rate their confidence.
 $0.02 are awarded for every correct
prediction and correct guess.

$40.00 bonus



In Phase two a $40.00 bonus is awarded to 3
individuals in two different groups.
In the first test group it is awarded to the subjects
with the highest number of correct post – prediction
classifications.
In the second group it was awarded to the subjects
with the highest rates of correct predictions.
“In the absence of selfsignaling the subject will
categorize the sign as male if,
and only if the probability
of male is greater than .5.
With
self - signaling, one has to
factor in the diagnostic utility
of selecting male.”


The subjects desire for
their prediction to be
accurate prompts them
to self deceive.
This deception makes
the subject less likely
to correctly confirm or
reject their prediction.
Application of the Formal Model
V(x=m, θO) -
V(x=f, θO)
>0
The total utility of a
confirmatory response
minus
The total utility of a
disconfirming response
Is great than zero

Stated simply:

A subject will choose a confirmatory response if there is
more total utility in a confirmatory response

We know there is no optimal outcome utility, so we must
examine the individual components of the equation
Application of the Formal Model
c(θOm - θOf) + λa(E(θm|x=m) - E(θm|x=f))
The reward for
choosing correctly *
(the deep belief that
the character is male
- the deep belief
that the character is
female)
Outcome Utility
The degree of self-deception * the reward for
correct anticipation * ((The experienced belief
that the character is male given that you
choose male) – (The experienced belief that
the character is male given that you choose
female))
Diagnostic Utility
What does this mean?
λa(E(θm|x=m) - E(θm|x=f))

Face-value


As mentioned earlier, in this system x implies θm, therefore,
self-deception occurs
Rational

Given rational discounting of (E(θm|x=m), the effect may
be diminished, but will always be positive. There will
always be confirmatory self-deception to some arbitrary
degree.
Results



The inconsistent responses give a baseline for subject
error.
There are proportionately more self deceptive
classifications than the error baseline.
The effect is greater in the anticipation group.
Results



Red: Anticipation
bonus
Green: Classification
bonus
At p=.001 27% of the
subjects in classification
group, and 45% of people
in the anticipation group are
self deceptive.
Results

Looking at individual changes in confidence
between phase one and two:
 Moderately
self deceptive individuals show an
increase in confidence following confirmation. +2.11
 High self deceptive individuals show a decrease in
confidence following disconfirmation. -1.76
 It is expected that a confirming response will
decrease confidence.
Results

Figure 3 shows that confirmation gives a confidence boost
at moderate rates.
Results

High self deception
correlates with fast
response time,
implying a
suppression of
evidence.
Discussion
Discussion


Self-deception can be induced when a large,
financial award is offered
People who have statistical bias achieve higher
confidence by self-deception – to a point
 Moderate
self-deception possibly related to increased
self-esteem and mental health


Many participants exhibited a statistical bias
This model can be applied to intrapersonal selfdeception as well as interpersonal deception
Two Agents



An actor
An observer
Evolutionary explanation
 It
is easier to deceive others in a mental state that is
ignorant of the individual’s true beliefs

Goal-setting argument
 Two
mental structures are required for successful goal
setting: one to choose which actions to take, one to evaluate
and reward the self for performance.
Alternative explanations

Perceptual bias
 Motivationally
biased perception of characters
 Does
not explain faster response time and is confounded by
randomized presentation of stimuli

Priming
 Exposure
to one’s prediction makes that gender more
salient during the selection phase
 Does
not explain effects seen in scaling of incentive
Applications

How far does the model extend?
 Confirmation
bias
 Denial
 Cognitive
 More?
dissonance
Download