Faculty Working Paper 93-0121
IH" LibKAi^Y OF THc
MAR 2 5
1993
URBAimm-oHAMPaiGN
Overconfidence in Initial Self-Efficacy Judgments:
Effects on Decision Processes and Performance
Dan
N. Stone
Department of Accountancy
University of Illinois
Bureau of Economic and Business Research
College of
Commerce and
Business Administration
University of Illinois at
Urbana-Champaign
BEBR
FACGLP/ WORKING PAPER NO. 93-0121
College of
Commerce and Business
University of
Illinois at
Administration
Grbana-Champaign
March 1993
Overconfidence in Initial Self- Efficacy Judgments:
Effects on Decision Processes and Performance
Dan
N. Stone
Department of Accountancy
Digitized by the Internet Archive
in
2011 with funding from
University of
Illinois
Urbana-Champaign
http://www.archive.org/details/overconfidencein93121ston
Overconfidence in Initial Self-Efficacy Judgments;
Effects on Decision Processes and Performance
by
Dan N. Stone
Accountancy and Information Systems
University of Illinois at Urbana-Champaign
Department
of
February, 1993
In
press, Organizational Behavior
and Human Decision Processes
Running Head: Overconfidence
in
Self-Efficacy
Overconfidence in Initial Self-Efficacy Judgments: Effects
on Decision Processes and Performance
Abstract
Judgments
of task-specific,
expected performance
one chooses
pursue and the extent
affect the activities
these
However,
activities.
efficacy
judgments or
performance
in
little is
initial,
of effort
how
devoted
to
of self-
on behavior, performance, and perceptions of
results of a pilot study
"first-impression" self-efficacy
complex tasks are biased towards overestimates
to illuminate
can
self-efficacy)
known about the accuracy
complex cognitive tasks. The
"overconfidence").
(i.e.,
relatively
their effects
experiment indicate that
cognitively
to
(i.e.,
and
judgments made
in
of personal ability
The experiment manipulated performance expectations
overestimates of
initial
self-efficacy affect decision processes,
performance, and perceptions of performance. Inducing positive expectations
produced overconfidence
in
choice accuracy, but did not increase
attention to strategy, or performance relative to mildly negative
negative expectations.
increased
In contrast,
effort,
and strongly
inducing mildly negative expectations
effort, attention to strategy,
and performance
relative to strongly
negative expectations. The results suggest that the demotivational effects of
initial
negative expectations are more robust than the motivational effects of
initial
positive expectations.
In addition,
may improve performance more
and
settings.
inducing mildly negative expectations
than positive expectations
in at
least
some
tasks
How
(i.e.,
accurate are
self-efficacy)?
and perceptions
of
initial
How does
judgments of task-specific, expected performance
perceived self-efficacy influence
performance? Social cognitive theories
effort,
(e.g.,
performance
Bandura, 1986;
Locke, 1991) have begun to explore the processes underlying the self-regulation
of behavior.
In
contrast to theories that posit unidirectional, deterministic effects
from environmental influences or internal dispositions on behavior, social
cognitive theories posit "triadic reciprocal causation" (Bandura, 1986),
behavioral, cognitive, and environmental influences interact
influence
one another. The
shift
which
in
and mutually
from deterministic, unidirectional theories
towards dynamic, bidirectional theories of sociocognitive functioning has
increased attention to self-referent processes (Locke, 1991). More recently,
social cognitive research
has focused on the linkages between judgments of
personal capability or "self-efficacy" and task performance. Judgments of selfefficacy are estimates of one's ability to attain a certain level of
specific task (Bandura, 1977, 1986).
Relatively
characteristics of self-efficacy judgments (Gist
&
little is
a
Mitchell,
1992) or their effects on
in
cognitively
complex
in press).
While many information sources
1977), comparisons with others are
affect self-efficacy
among
(Bandura, 1986; Festinger, 1954; Goethals
activities
judgments (Bandura,
the most important influences
&
Such comparisons are a primary influence on
most human
in
known about the
behavior, performance, and perceptions of performance
tasks (Cervone,
performance
Darley, 1977; Suls
self-efficacy
&
Miller,
1977).
judgments because
do not provide objective, nonsocial evidence of
performance (Bandura & Jourden, 1991). People therefore often gauge
expected and actual performance by comparison with that of others.
their
Overconfidence
in
Self-Efficacy
2
Overconfidence and Judgments of Self-Efficacy
Judgments
are subject to the
capacity,
same
and memory
self-efficacy
same
of self-efficacy are
made under
uncertainty.
As a
result,
they
constraints of limited attention, information processing
that characterize other decisions.
Research indicates
judgments are products of the same heuristics and subject
biases as other judgments under uncertainty (Cervone
Peake & Cervone, 1989; Switzer & Sniezek,
& Peake,
that
to the
1986;
1991). While research has
established that self-efficacy judgments are the product of heuristic-based
processes, relatively
little is
However, there
is
known about
an extensive
the accuracy of self-efficacy judgments.
literature dating
from the 1940s
(e.g.,
Festinger, 1942; Frank, 1953; Irwin, 1944) that suggests a tendency towards
overly positive self-evaluations.
The
bias towards overly positive self-evaluations
extends across an impressive range of research paradigms, tasks, and
participants (see Greenwald, 1980,
Evidence of
this
and Taylor & Brown, 1988,
tendency includes recalling positive personality
than negative ones
(e.g.,
Kuiper, dinger,
one's self more positively than others
illusions of control
over chance events
One
Many
traits
more
easily
MacDonald & Shaw, 1985), evaluating
(e.g.,
Green & Gross, 1979),
(e.g.,
optimism about future task performance
1955).
for reviews).
unrealistic
Langer, 1975), and unrealistic
(e.g.,
Crandall, Solomon,
&
Kelleway,
recent review summarizes this literature as follows:
researchers have studied biases in the processing of selfrelevant information and have given their similar phenomena
different names. There is, however, considerable overlap in
findings, and three that consistently emerge can be labeled
unrealistically positive views of the self, exaggerated perceptions of
personal control, and unrealistic optimism (italics in original)
(Taylor & Brown, 1988, p. 194).
Overconfidence
Overly positive self-evaluations also appear to be present
one's performance
than objective evidence indicates
Fischhoff, 1980;
in
&
Lichtenstein, 1977; Koriat, Lichtenstein,
Paese & Sniezek, 1991),
&
identifying
Birdsall, 1961), predicting
and diagnosing the malignancy
"Overconfidence" occurs
performance exceed
when
horse races (Fischhoff
of ulcers (Fischhoff
Slovic, 1980).
performance. Explanations for overconfidence
&
Hogarth, 1978)
seek confirming, but not disconfirming, evidence (Hoch, 1985;
to
Klayman&Ha,
1987).
Overconfidence
Lichtenstein, Fischhoff,
is
a relatively robust
&
Phillips,
phenomenon (see Keren,
in
which overconfidence
even eliminated. For example, certain types
experience appear to be useful
in
1
991 and
,
1982, for reviews), although recent research
suggests the existence of conditions
Romney &
&
&
decision makers' beliefs about the quality of their
their actual
and a tendency
in
&
words and sounds (Clarke,
include the structural characteristics of judgment tasks (Einhorn
forecasters
evaluations of
in
a variety of tasks, including answering
(Fischhoff, Slovic,
1960; Swets, Tanner,
Slovic, 1980),
3
Self-Efficacy
decision tasks. Decision makers believe they perform better
in
almanac questions
in
is
lessened and
of professional training
eliminating overconfidence (e.g.,
Murphy & Winkler, 1977;
and
weather
financial auditors in Tomassini,
Solomon,
Krogstad, 1982). Similarly, overconfidence appears to be either
greatly reduced (Sniezek
Kleinbolting, 1991)
when
&
Buckley, 1991) or eliminated (Gigerenzer, Hoffrage,
decision makers evaluate their performance over a
series of decisions rather than making evaluations of
Buckley (1991
)
decisions result
performance
argue that
in
&
this
each decision. Sniezek and
occurs because evaluations of a series of
greater weighting of self-evaluative reactions to one's
relative to other factors (e.g., self-efficacy).
Overconfidence
There
prior to
having first-hand task experience,
example, Cervone and
Wood
ability relative to
Schunk (1981) treated
programs
may
4
made
overconfidence. For
reflect
(1992) found that participants' self-efficacy
judgments made before engaging
different
Self-Efficacy
evidence that self-efficacy judgments, particularly those
is
overestimated
in
in
a complex decision making task significantly
subsequent performance.
children with gross deficiencies
of self-directed learning.
in
Similarly,
Bandura and
mathematics using four
Before treatment,
all
four groups
overestimated the number of subtraction problems they could solve. After
receiving treatment and extensive performance feedback, three of the four groups
still
significantly overestimated their capability at
even the pre-treatment
self-efficacy
overestimates of
(See Figure
ability
mathematics problems. Indeed,
judgments of severe phobics
1
in
reflect
Bandura & Adams, 1977, and Figure 2
in
Bandura, Adams, Hardy, & Howells, 1980).
Although there
judgments, relatively
made
is
some evidence regarding
little
attention
has been given
prior to having first-hand task experience.
judgments of
choose
to
the accuracy of self-efficacy
"first-impression"
Initial
self-efficacy are important, since they
pursue (Bandura, 1986), the extent
judgments of self-efficacy
to
can
of effort
affect the activities
devoted
to
these
(Cervone & Peake, 1986; Cervone & Palmer, 1990), and subsequent
judgments (Cervone & Palmer, 1990). One procedure
overconfidence
in self-efficacy
judgments
is
to
incorporates explicit comparisons with others
75th percentile"). "
underestimates of
50th percentile.
If
for
people
activities
self-efficacy
measuring
use a percentile rank measure that
(e.g.,
"I
think
I
will
perform at the
there are no systematic biases towards over or
ability,
percentile rank self-efficacy judgments should
In contrast,
if
initial
self-efficacy
judgments
be
at the
reflect
overconfidence, percentile rank self-efficacy judgments should exceed the 50th
Overconfidence
percentile.
efficacy
The
pilot
judgments
study tests the hypothesis that
reflect
overestimates of
initial
in
Self-Efficacy
5
percentile rank self-
ability.
Study
Pilot
Method
Undergraduate students enrolled
Participants.
in
an undergraduate
business course (n=47) participated for course credit and the opportunity to win a
randomly distributed $20 cash
the study
was
to
Participants
prize. 2
were
purpose of
told that the
understand better how students choose colleges. They then
read a case describing a student interested
in
choosing a college. The case
described six attributes that the student had determined to be important (student
quality, student/faculty ratio, distance
from home,
total cost,
percentage of faculty with Ph.D.s). Participants were
[The student] believes
that,
while
all
school size and
told that:
the factors are relevant
in
making the decision, some factors are more important than others.
Specifically, he thinks that factors one and two are three times as
important as factors five and six, while factors three and four are
twice as important as factors five and six.
Participants
were
told that the task consisted of
choosing the college that
best matched these criteria from a set of eight alternatives.
The experimenter
then showed an example of a multiattribute choice display using a microcomputer
and the Mouselab decision research software (Johnson, Payne, Schkade, &
Bettman, 1989). Following the demonstration, participants asked clarifying
questions about the task and the
criteria.
They then
choice accuracy percentile rank on the task
enrolled in the class
and
(2)
how much
they had seen would help them
in
in
stated: (1
)
their
expected
comparison with other students
they thought the hardware and software
making choices. Participants stated
expected percentile rank as a number between
their
and 100 and the extent
of help
Overconfidence
in
Self-Efficacy
6
they expected from hardware and software on a 7 point Likert-type scale. Before
asking these questions, the experimenter explained the meaning of percentile
ranks and emphasized
at or
by
that,
definition, half of the participants
should be ranked
below the 50th percentile.
Results and Discussion
-
The mean expected
which
is
Pilot
Study
percentile rank
was 69.6 (standard
significantly higher than the 50th percentile (f(46)
deviation, 14.9),
= 8.92, p < .001).
Thirty-seven participants (78.7%) expected to perform above the 50th percentile,
seven (14.9%) expected
expected
to
significantly
to
perform at the 50th percentile, and three (6.4%)
perform below the 50th percentile.
more
A
sign test (Hays, 1981) indicates
participants expected to perform
percentile (z = 66.0,
p<
.001
).
above than below the 50th
Given the impossibility of
79%
of the participants
performing above the 50th percentile, the results provide support for the
hypothesis that
reflect
initial
judgments
of self-efficacy in cognitively
complex tasks
overconfidence.
Self-Efficacy, Behavior,
and Performance
in Multiattribute
Choice
Social-cognitive theory posits three self-referent processes that potentially
influence behavior
and performance:
self-set goals, self-evaluative reactions,
and
judgments of self-efficacy (Bandura & Cervone, 1983, 1986). Setting personal
goals can influence behavior by providing a benchmark for evaluating one's
performance. Self-evaluative reactions are feelings of satisfaction or
dissatisfaction with one's performance.
Self-evaluative reactions influence
performance and behavior primarily when one both sets personal goals and
receives feedback about performance (Bandura
in
& Cervone,
1983, 1986; Cervone,
press). This suggests that the influence of self-efficacy relative to self-
Overconfldence
evaluative reactions
feedback and
is likely
to
be greatest
in
in
the absence of performance
self-set goals.
Existing research has studied the effects of self-efficacy judgments
performance
in
Jourden, 1992;
in
&
&
a complex decision making task (Bandura
Bandura & Wood, 1989; Cervone,
Switzer
7
Self-Efficacy
Wood &
on
Jourden, 1991;
& Wood, 1991; Cer/one & Wood,
Jiwani,
Bandura, 1989). However with few exceptions
Sniezek, 1991), research to date has studied the
same
1992;
(e.g.,
task and setting,
which participants manage the employees of a simulated manufacturing
business. Studying alternative tasks and settings provides important evidence on
the robustness of self-efficacy effects
Wood,
complex decision making (Cervone &
in
1992).
Choosing among alternatives
is
one
of the
most pervasive, and
challenging, of cognitive activities (Hogarth, 1983; Stone
Multiattribute choice (e.g., Payne, 1976)
demands both
effort (i.e.,
decision strategies
1991).
a cognitively complex task that
working hard) and the development of appropriate
working smart).
(i.e.,
is
& Schkade,
In
such tasks, alternatives are frequently
displayed on one dimension of a matrix and attributes of alternatives on the other.
For example
in
Figure
1,
colleges (labeled "A," "B," "C," etc.) are displayed as
columns, and the attributes of the colleges
Ratio," etc.) as rows,
in
a matrix.
(i.e.,
"Student Quality," "Student/fac.
One advantage
efficacy in multiattribute choice tasks
is
of studying the effects of self-
the existence of well-defined measures of
decision processes (Klayman, 1983; Payne, 1976; Payne, Braunstein,
&
Carroll,
1978).
Insert Figure
An
1
about here
important question related to self-efficacy
is
the effect of positive and
negative information about one's expected performance on self-efficacy
Overconfidence
judgments (Bandura & Jourden, 1991). More
efficacy judgments of participants
manipulated
differ
how
specifically,
in
Self-Efficacy
might the
self-
whose performance expectations are
from the self-efficacy judgments of participants
in
who expect
the
pilot
study? Will the increases
in self-efficacy of
performance than the
study participants be larger than the corresponding
decreases
pilot
in self-efficacy
8
of participants
participants
better
who expect worse performance
than
pilot
study participants? There are reasons to believe that positive information about
expected performance
information.
will
have larger
For example, there
is
1985).
It
may
&
on
self-efficacy than negative
evidence that people process and
positive information about themselves
information (Kuiper
effects
more
Derry, 1982; Kuiper
and
easily
efficiently
recall
than negative
& MacDonald, 1982; Kuiper
et
al.,
therefore be the case that positive information about one's
expected performance leads
to larger increases in self-efficacy
than the
corresponding decreases associated with negative information about expected
performance.
Historically, self-efficacy theory posited that
would perform a task
better than those
well exerted
who
Eden, 1990). As a
more
effort,
thought that they would
result,
people
who
believed that they
persevered longer, and performed
fail
(Bandura, 1977, 1982, 1986;
researchers generally argued for the existence of
symmetric performance effects from positive and negative self-expectations.
Positive self-expectations
were hypothesized
to increase
performance, and
negative self-expectations to decrease performance, through mediating effects on
effort
and perseverance.
However, most of the evidence supporting a symmetric relationship
between expectations and performance examined
solving anagrams, simple addition, typing).
the effects of self-referent processes
in
cognitively simple tasks (e.g.,
More recent research suggests
cognitively
complex tasks may
that
Overconfidence
qualitatively differ from those
Wood,
1992).
In
found
in
simple tasks (Cervone,
in
in
Self-Efficacy
press;
cognitively simple tasks, immediate performance
Cervone &
feedback
confirms the existence of an unequivocal, positive relationship between
performance (Bandura & Jourden, 1991). As a
9
effort
and
unequivocal
result of this
effort/performance relationship, positive expectations induce greater effort that
immediately and directly improves performance.
cognitively
and smart
In contrast,
complex tasks depends upon both working hard
(I.e.,
attention to strategy development).
In
therefore
change the influence
and performance
There
is
in
cognitively
some evidence
(Clarke, 1960; Nickerson
therefore be larger
in
that
cognitively
in
if
tasks,
immediate
overconfidence increases with task
Pitz,
difficulty
1974). Overconfidence
complex than simple
tasks.
If
may
judgments of
self-
complex cognitive tasks, decision makers who
effort or attention to strategy
tasks that lack immediate, unequivocal performance feedback
work hard
complex
processes on decision processes
expect to perform well are unlikely to increase their
in
into
effort)
complex tasks.
& McGoldrick, 1965;
efficacy overestimate ability
greater
in
performance feedback may
direct
of self-referent
(i.e.,
cognitively
working harder and smarter does not necessarily translate
performance improvements. The lack of
performance
.
After
all,
why
you expect superior performance and receive no feedback that
contradicts this expectation
that induces positive
(cf.
Bandura & Jourden, 1991)? Providing information
performance expectations
post-decision perceptions
of,
is
therefore likely to increase
but not actual, performance relative to providing
information that induces mildly negative and strongly negative performance
expectations.
Inducing positive expectations should therefore increase
overconfidence but not task performance.
Research by Sniezek and colleagues provides support
that positive
for the
argument
performance expectations may induce overconfidence without
Overconfidence
concomitant increases
led decision
makers
or performance. Trafimow
in effort
to believe they
would either perform
in
Self-Efficacy
and Sniezek
10
press)
(in
well or poorly in
answering general knowledge questions. Participants who believed they would
perform well did no better at answering questions but were significantly
overconfident.
In contrast,
participants
who expected
neither over nor underconfident. Switzer
engaged
participants
expectations.
correctly transferred
perform poorly were
and Sniezek (1991) provided
a text editing task with either high or low performance
Participants
would perform better
participants
in
to
who
at the task
received high performance anchors believed they
and exert more
effort.
However, they neither
more sentences nor executed more keystrokes than
who expected
to perform poorly.
Information that induces mildly negative expectations
"average" performance)
may
increase
effort,
attention to strategy,
performance by increasing the perceived challenge
1990; Csikzentmihaiyi
&
LeFevre, 1989).
In
(e.g.,
of tasks
expecting only
and
Csikzentmihaiyi,
(cf.
contrast to the overconfidence
induced by positive expectations, mildly negative expectations may increase the
perceived importance of exerting
effort
and attending
to strategy
development.
However, strongly negative performance expectations may decrease the
perceived benefits of exerting
at
a task
in
which one expects
effort,
since there
is little
benefit from working hard
to perform poorly regardless of one's effort
Bandura & Jourden, 1991; Sarason, 1975; Sarason, Sarason &
(cf.
Pierce, 1990).
Mildly negative expectations should therefore improve task performance by
Increasing effort exerted and attention to strategy, while strongly negative
expectations should decrease task performance by decreasing
effort
and
attention
to strategy.
Research suggests
that inducing negative
performance expectations can
reduce and even eliminate overconfidence (Trafimow & Sniezek,
in
press).
Mildly
Overconfidence
in
Self-Efficacy
negative expectations
may
outperform their
expectations as the result of exerting effort at the task.
initial
eliminate overconfidence
Strongly negative expectations
self-fulfilling effects of
may
negative performance expectations on
in
may
performance that matches one's
An experiment was designed
increase
because decision makers
eliminate overconfidence as the result of the
specifically, strongly negative expectations
results in low
will
More
effort.
lead to decreased effort which
initial
low expectations.
to test the following
hypotheses:
self-efficacy of participants given positive information
performance
11
be larger than the corresponding decrease
in
(1)
the
about expected
self-efficacy of
participants given negative information about expected performance, (2)
participants given positive expectations about their performance
perceived performance than participants
negative performance expectations,
performance expectations
and, as a result,
will
make more
have higher
are given mildly and strongly
(3) participants
given mildly negative
use more effective strategies, exert more
effort,
accurate choices than participants given strongly
negative performance expectations and,
performance expectations
who
will
will
(4) participants
be overconfident
(i.e.,
given positive
perceived
will
exceed actual
performance), while participants given mildly and strongly negative performance
expectations
will
be neither over nor underconfident.
Experiment
Method
Procedure. The experiment took place one
week
after the pilot study.
Undergraduate students from the same class and semester as the
pilot
(n=139) participated for course credit and the opportunity to win one of
study
five
Overconfidence
randomly distributed $20 cash
participated
Initial
in
the
None
prizes.
in
12
Self-Efficacy
of the participants in the experiment
pilot study.
procedures were identical
to
those described
in
the
pilot
study
through the demonstration of the multiattribute choice display. Following
this,
participants listened to a 10 minute presentation by the author designed to
manipulate their expected task performance. They then responded
two questions as participants
in
the pilot study.
These questions asked:
expected choice accuracy percentile rank on the task
students enrolled
the class and (2)
in
software would help them
in
to the
how much
in
same
(1) their
comparison with other
they thought the hardware and
making choices.
Subsequently, participants moved to a different room where they were
greeted by a proctor
who was unaware
of the
purpose of the experiment. This
proctor seated participants at computer terminals.
choices of colleges
alternatives
practice
(1
colleges).
(i.e.,
and 12
Following
Participants then
actual) from choice sets containing 8
this,
participants
completed a post-
experimental questionnaire. Participants completed the task
Each session
required,
made 13
in
groups of 8
to 13.
on average, 50 minutes.
Design. Participants were randomly assigned to one of three expectation
conditions.
In
the positive expectation condition participants were told that the
,
software they would use to choose colleges
system designed
to
Improve the accuracy of
a result of using the software,
better than
90%
if
was an advanced
their choices.
They were
told that
expectation condition participants were told that
In
the mildlv negative
some
of the other participants
would be using an advanced decision support system designed
However they would use software
to
improve the
was
accuracy of
their choices.
designed
improve decision making, "because the advanced decision support
to
as
they worked hard, their choice accuracy would be
of previous research participants.
,
decision support
that
not
Overconfidence
system
is
They were also
not available to us today."
they would do better than
50%
in
Self-Efficacy
13
by working hard,
told that,
of previous research participants.
In
the strongly
negative expectation condition participants were told that most of the other
,
participants
would be using an advanced decision support system designed
improve the accuracy of
was
their choices.
However, they would use software that
not designed to improve decision making, "because the
support system
is
hard, they would
do better than 10%
In reality, all participants
advanced decision
They were also
not available to us today."
still
to
told that,
by working
of previous research participants.
used the Mouselab decision research software,
which collects data on decision processes. The software did not provide any
decision supportive capabilities. Because the experimental manipulation involved
deception, participants were
told,
and given an opportunity
to
ask questions
about, the true purpose of the experiment and the software at the end of the
semester.
The
levels of percentile ranks
those given by participants
in
used
for the manipulation
The
the pilot study.
positive
were based upon
and
mildly negative
expectation condition levels (90th and 50th percentile) were approximately
equidistant from the average self-efficacy of the pilot study participants (69.6).
The
strongly negative expectation condition level (10th percentile) provided a
symmetric percentile rank
relative to the difference
mildly negative conditions
(i.e.,
A
differing
potential
90
-
50 = 50
-
between the
positive
and
10).
problem with manipulating performance expectations
is
that
performance expectation might invoke self-evaluative reactions among
participants.
feedback
that
For example,
if
participants expected to perform well but received
suggested they had performed poorly, then dissatisfaction with
performance could influence subsequent behavior and performance. Social
cognitive theory posits that both personal goals
and knowledge
of one's
Overconfidence
performance are necessarily
in press).
were not asked
were not provided with feedback
procedure intended
participants
were
external reason
at
To lessen
told they could
to state personal
performance goals and
skills, etc.) (cf.
Data
Collection.
Trafimow and Sniezek,
"hidden"
the
in
mouse
time.
first
was
that
was provided
in
to
them) and
(e.g., intelligence,
press).
Using Mouselab, alternatives are displayed as one
dimension of a matrix, and
choice set
second
expect to do well or poorly based upon an
not based upon their personal abilities or characteristics
computer
A
any time during the experiment.
the quality of the software that
(i.e.,
effects from self-evaluative
lessen effects from self-evaluative reactions
to
14
(Bandura &
to fully activate self-evaluative reactions
Cervone, 1983, 1986; Cervone,
reactions, participants
Self-Efficacy
in
their attributes
on the other (see Figure
1 ).
When
appears on the screen, the information about the alternatives
a
is
boxes. These boxes can be "opened" to reveal their contents by using
to
move
Participants
the cursor into a given box. Only
make choices by moving
desired alternative and clicking a
mouse
one box can be open
at a
the cursor to the choice box of the
button.
Figure
1
illustrates the
experimental display after a box has been "opened." The data collected using
Mouselab includes the sequence
and the choice made by a
Two dimensions
of
boxes opened, the time spent
in
each
box,
participant.
of the information display
were counterbalanced: the
presentation order of the twelve choice sets and the order of alternatives within a
choice
set.
Counterbalancing was between-participants.
A fractional
factorial
design was used to select combinations for counterbalanced factors (Hays, 1981).
Choice Accuracy. The alternatives
best to worst using the criteria given
in
in
each choice set were ranked from
the case. For example, the best
alternative in a choice set received a rank of 8, the
the worst a rank of
1
.
second best a rank of 7, and
Absolute choice accuracy was computed as the average
Overconfidence
in
15
Self-Efficacy
choice quality ranking of each participant's chosen alternatives across the 12
choices. This
measure was then converted
to
a percentile rank relative to other
participants (e.g., average absolute choice accuracy of 6.5 equaled the 58th
Data analysis used the percentile rank measure
percentile).
comparisons with
participants'
to permit
expected and post-experimental perceived choice
accuracy.
Decision Process Measures.
Two broad
indicators of decision strategies
are measures of information search and measures of cognitive effort (Payne et
al.,
1978).
Two measures
variability in time
attribute
spent per alternative and
(2) variability in
(Klayman, 1983; Payne, 1976). The
alternative (attribute)
few alternatives
measures the extent
(attributes)
to
make
accuracy of self-efficacy judgments relative
The experiment measured
alternatives (attributes).
effort
pilot
was measured as the
study measured the
to the characteristics of percentile
in
the pilot study), and (2) by comparing
judgments with actual task performance by
Debriefing Questionnaire.
Upon completing
stated: (1) their choice accuracy percentile rank
in
is
(1) relative to the characteristics of
aggregated percentile rank measures (as
other students enrolled
It
the accuracy of both pre- and post-
experimental self-efficacy judgments:
self-efficacy
focused on a
a choice.
Accuracy of Self-Efficacy Judgments. The
ranks.
is
of the time spent per alternative (attribute)
across the set of attributes (alternatives). Cognitive
time required to
spent per
which decision time
all
the: (1)
time spent per
variability in time
versus spread across
computed as the standard deviation
total
search are
of the selectively of information
individuals.
the experiment, participants
on the task
in
comparison with
the class and (2) the extent to which the hardware and
software they used helped their decision making.
Overconfidence
Analysis. Analysis of variance
(ANOVA) and
evaluate the data. Post-hoc comparisons were
two-tailed
made
t
in
tests
Self-Efficacy
16
were used
to
using the Tukey
HSD
test
(Hays. 1981).
Results and Discussion
Manipulation Checks. Significant differences exist
percentile ranks of
three groups (see row
all
1
of
Table
percentile rank of the positive expectations group
was
group, 56.3, and the strongly negative group, 31.3.
differences
in
1
in
).
the expected
The expected
88.8, the mildly negative
The
direction of the
expected percentile ranks were therefore consistent with the
experimental manipulation. The expected percentile rank of participants
positive condition
(i.e.,
was
in
the experimental manipulation (f(46) = -0.91,
.19).
However, the expected percentile ranks of participants
2.31,
p=
the values
{t{46) = 5.85,
and strongly negative
(i.e.,
50 and 10) provided
Insert
in
1
p < .001) conditions exceeded
in
participants' beliefs about
in
how much
the experimental task (row 2 of
Consistent with the experimental manipulation, participants
positive expectations
in
the
group believed they would receive more help than
participants in the mildly
and strongly negative conditions.
Comparison of Self-Efficacy Judgments
hypothesis states that the increase
in
with Pilot Study.
The
first
self-efficacy of participants given positive
information about expected performance
decrease
the mildly (^(44) =
about here
the hardware and software would help them
1).
in
p=
the experimental manipulation.
Table
Significant differences also exist
Table
the
not significantly different from the percentile rank value
90th percentile) provided
.03)
in
in self-efficacy of participants
will
be larger than the corresponding
given negative information about
Overconfidence
expected performance. The
fact that the self-efficacy
judgments of participants
the positive expectations group were equivalent to the manipulated value
90th percentile), while the self-efficacy judgments of participants
negative expectations group exceeded the manipulated value
percentile) provides
To more
group.
rank for the
some support for
study participants)
(i.e.,
the mildly
50th
(i.e.,
mean expected
was subtracted from the expected
percentile
percentile
ranks of the positive and mildly negative expectation condition participants
The
experiment.
sign of this
measure was then reversed
mildly negative expectations condition.
changes
in
in
larger increases in the positive expectations
formally test the hypothesis, 69.6 (the
pilot
in
17
Self-Efficacy
in
Positive
in
the
for participants in the
numbers therefore
the direction of the experimental manipulations,
i.e.,
indicate
> 69.6 for positive
expectation condition participants and < 69.6 for mildly negative expectation
participants.
Negative numbers indicate changes
the experimental manipulations,
participants
and > 69.6
The changes
in self-efficacy
The average change
was
< 69.6 for positive expectation condition
in
of participants in the positive condition
in self-efficacy in
number
in self-efficacy
it
the positive expectation condition
was
of participants in the positive
significantly
more
13.3.
To
better understand
of positive condition participants
and
were
(1
24.4%) whose expectations were greater than 69.6 than participants
49.1
,
p<
.01
).
The
whose
was computed (see Table
participants in the mildly negative condition
positive condition (2/47 =
why
larger, the
mildly negative conditions
expectations were greater and less than 69.6
were
were
the mildly negative condition (f(91) = 1.98, p =
19.2, in the mildly negative condition
the changes
the opposite direction from
for strongly negative expectation participants.
larger than those of participants
.05).
i.e.,
in
2).
There
1/45 =
in
the
4.3%) whose expectations were less than 69.6 (X^ =
results therefore support the hypothesis that, relative to the
expectations of participants
in
the pilot study, the increase
in self-efficacy in
the
Overconfidence
positive expectation condition
exceeds the decrease
in
Self-Efficacy
in self-efficacy in
18
the mildly
negative condition.
Insert
Table 2 about here
Choice Accuracy and Cognitive
expectation condition
participants
Table
1 ).
measures
in
made
Effort.
better choices
Participants
and took longer
the mildly negative
choose than
to
the strongly negative expectation condition (see rows 3 and 4 of
Positive expectation condition participants
that
in
were not
mean accuracy
significantly different
was 56.4
percentile rank
had accuracy and
effort
from the other conditions. The
for mildly negative condition participants,
52.7 for positive condition participants, and 41.1 for strongly negative condition
participants.
Participants
126.1 seconds to
make
in
the mildly negative expectations condition averaged
choices, positive expectation participants averaged 111.5
seconds, while strongly negative expectation participants averaged 87.9 seconds.
Information Search. Higher variability
p<
.01
)
and per
attribute (r
=
accuracy. Greater selectivity
.22,
in
p<
)
was
time spent per alternative (r= .22,
positively correlated with choice
information search therefore
higher choice accuracy. Participants
variability in time
.01
in
in
spent per alternative and per attribute than participants
in
effort,
the variability
the positive expectation condition
participants
in
with
the mildly negative condition had higher
strongly negative condition (see rows 5 and 6 of Table
performance and
was associated
in
was
1 ).
As
with
in
the
measures
of
time spent per alternative of participants
not significantly different from that of
the other conditions.^
Post-experimental Measures. Consistent with previous research
demonstrating the persistence of expectations (Cervone
&
Palmer, 1990), post-
experimental measures of perception closely followed expectations. Significant
Overconfidence
differences exist
(row 7 of Table
the perceived accuracy percentile ranks of
in
all
in
19
three groups
with the positive expectations group perceiving the highest
1 ),
percentile rank (80th), the mildly strongly negative group next (55th)
strongly negative group last (32nd).
exist in the
Self-Efficacy
amount
and the
Differences consistent with expectations also
of help participants believed they received from the
and software they used (row 8
of
Table
1
Participants
).
in
hardware
the positive
expectations group continued to believe they received more help from the
hardware and software than participants
the mildly negative and strongly
in
negative conditions.
Accuracy of Self-Efficacy Judgments. The mean expected and postexperimental percentile ranks across conditions were 58.8 and 55.4, respectively.
Both the expected {t{^38) = 3.42, p < .001) and post-experimental {t{^38) = 2.33,
p=
.02) percentile ranks
were
significantly higher than the 50th percentile,
both expected and post-
indicating significant overall overconfidence
in
experimental self-efficacy judgments. There
was no
overall self-efficacy
(f(138) = 0.12,
p=
significant
vs. actual (f(44)
,
percentile rank
in
)).
The data
vs. actual (f(46)
perceived (f(46) =
underconfidence
-1 .72,
in
exceeded
post-experimental
therefore support the existence of
In contrast,
the actual
marginally greater than the expected and post-experimental
perceived percentile ranks for participants
(expected
vs.
the positive expectation condition.
was
percentile ranks
the positive expectation condition
= 7.89, p < .001), actual
perceived (f(44) = 5.91 p < .001
in
mean
.91).
the actual percentile rank for participants
overconfidence
in
between the expected and post-experimental judgments
The expected and post-experimental perceived
(expected
change
= -1.82, p =
p=
.10).
in
the strongly negative condition
.08), actual vs.
The data
post-experimental
therefore provide
weak evidence
the strongly negative expectation condition. There
of
were no
.
Overconfidence
significant differences
between expected and actual performance
= .97) or between actual and perceived performance
participants
among
p <
.01),
= .17, p =
.05).
=
.28).
.09,
p=
(f(46)
= -0.03, p
= -0.40, p = .66) for
Expected, Actual and Perceived Performance.
Expected and perceived performance were
.82,
as were,
significantly
to a lesser extent,
In contrast,
and
positively correlated
perceived and actual performance
cognitive effort
(r
expected and actual performance were uncorrelated
Decision Processes as Mediating Variables. To what extent do changes
for the
20
Self-Efficacy
the mildly negative condition.
in
Relationships
(r=
(f(46)
in
(r
in
and decision processes between the experimental groups account
choice accuracy results? To demonstrate a mediating relationship
between the expectation manipulation, cognitive processes, and decision
performance,
it
is
necessary
to establish three conditions: (1) that the expectation
manipulation affects processing,
(2) that
the effort and processing measures are
correlated with accuracy, and (3) that the effect of expectations on choice
accuracy
is
weakened
covariates (Baron
or eliminated
& Kenney,
if
processing measures are used as
1986).
The previous analyses demonstrate
affects both cognitive effort
Measures
p <
with choice accuracy,
to
and by
p<
.01
),
and
satisfying condition
variability in
p <
attributes (r= .22,
condition 2.
ANCOVAs with
choice and the
correlated, only
The
.01)
fulfilling
examined by computing
Since time
and decision processes,
of time to choice (r = .36,
alternatives (r= .22,
that the expectation manipulation
The
1
search by
.01),
are correlated
third criterion for
mediation
was
processing measures as covariates.
variability in
search measures were highly
one processing measure was used
effect of the expectation manipulation
was
time to choice as a covariate (F(2,136) = 1.3,
in
each
of three
not significant
p=
.27).
in
ANCOVAs.
the presence of
Thus, the effect of
Overconfidence
expectations on choice accuracy
effort
expended under
.08), variability in
accounted
is
significantly
(variability in time
weakened
in
summary, the
in
by the
The
Self-Efficacy
effect of the
the presence of information
time spent per attribute (F(2,136) = 2.7, p =
partially
is
21
differential cognitive
spent per alternative (F(2,136) =
expectations on decision performance therefore
differences
for
differing expectation conditions.
expectation manipulation
search covariates
is fully
in
.07)).
2.5,
p=
The impact
accounted
for
information search under differing expectation conditions.
of
by the
In
results indicate that expectations affected choice accuracy through
the mediating influences of cognitive effort and information search.
General Discussion
This
is
the
first
study to demonstrate that self-referent thought can affect
decision processes and performance
initial
in
self-efficacy affected the decision
affected decision performance.
provide evidence
in
a
multiattribute choice tasks.
Changes
processes of participants, which
in turn,
Further, the results of the mediation analysis
new task and
context that the effects of self-efficacy on
performance operate primarily through the mediating influences of cognitive
and information search. The
processes used
research.
The
in this
task,
research
in
and the measures
differ substantially
results therefore provide important
of self-efficacy
from those used
effort
and decision
in
previous
evidence on the robustness of
both self-efficacy influences on decision making, and on the role of effort and
strategy
in
The
cognitively
mediating the relationship between self-efficacy and performance.
results of both the pilot study
complex tasks
overconfidence.
participants
In
the
and experiment indicate
that lack feedback, self-efficacy
pilot study,
were overconfident
the
initial
self-efficacy
that in
judgments tend towards
judgments of
relative to the defined characteristics of percentile
Overconfidence
ranks.
the experiment, both
In
efficacy
were overconfident
initial
in
Self-Efficacy
and post-experimental measures
22
of self-
relative to the characteristics of percentile ranks.
In
addition, relative to the self-efficacy of participants in the pilot study, the increase
the positive expectation condition participants
in self-efficacy of
exceeded the decrease
participants,
self-efficacy.
performance
in self-efficacy of
to
decreases than increases
However, overconfidence was eliminated
relative to actual
low self-efficacy compared with participants
Interestingly, although
overconfidence was eliminated
and strongly negative expectation
negative condition.
One problem observed
reduce overconfidence
(e.g..
is
in
in
the pilot study.
both the mildly
in
the strongly
can also decrease task
Trafimow & Sniezek,
negative expectations condition replicates
in
previous manipulations intended to
that reducing overconfidence
Experiment 2
in
conditions, the task performance of participants
the mildly negative condition exceeded that of participants
performance
in
the mildly and strongly negative conditions of the experiment by
explicitly inducing
in
the experiment
the mildly negative condition
which suggests greater resistance
in
in
this result.
in
In
press).
strongly
the strongly negative
expectations condition, overconfidence, task performance, and
relative to the mildly negative expectations condition.
The
effort
decreased
Reducing overconfidence
without simultaneously reducing task performance therefore appears to be a
gentle
More
art.
specifically,
performance expectations must be decreased
without inducing the sometimes accompanying self-defeating belief that effort
is
irrelevant to performance.
The
finding of overconfidence in self-efficacy
judgments provides additional
evidence supporting the argument that overly positive self-evaluations are
relatively
common
however,
prior to
in
human
cognition (Taylor
Cervone and
Wood
addressed the issue had observed
& Brown,
(1992), research
1988).
Interestingly
in self-efficacy
that self-evaluations
were
that
relatively
accurate
Overconfidence
(e.g.,
Bandura & Adams, 1977; Bandura
judgments
of accurate self-efficacy
in
et
al.,
One reason
1980).
early self-efficacy research
Self-Efficacy
in
23
for the finding
may be
that in
the few studies that examined the accuracy of self-efficacy judgments, participants
were severe phobics who judged
handling).
It
may be
self-efficacy in phobia-relevant tasks
the case that phobics' self-efficacy judgments
related tasks are relatively accurate, while those of nonphobics
snake
phobic-
in
(i.e.,
(i.e.,
most people)
often overestimate self-capability.
In
in
the experiment reported herein, there
self-efficacy
perhaps
if
between the
participants
initial
in
change
and post-experimental judgments. However,
of their judgments,
overconfidence would have been eliminated. Cervone
and Palmer (1990) studied the
engaged
significant overall
had received feedback on the accuracy
at least post-experimental
participants
was no
effects of
feedback on the efficacy judgments of
a problem-solving exercise
in
which
all
received performance feedback. The experimenters provided
participants
some
participants
with randomly assigned performance expectations, other participants did not
receive experimenter-assigned performance expectations.
judgments predicted subsequent levels of
Initial
self-efficacy
self-efficacy for participants
received randomly assigned performance expectations. However,
subsequent
self-efficacy
judgments were uncorrelated
who
initial
for participants
who
receive experimenter-assigned performance expectations. Participants
not receive experimenter-assigned expectations therefore
available feedback. While Cervone
and Palmer
one has
in
little
confidence
in
in
did not
who
greater use of
their findings is that
improving the accuracy of self-efficacy judgments,
one's
one's self-efficacy judgment
is
initial
did
did not explicitly study the
accuracy of self-efficacy judgments, one interpretation of
feedback may be useful
made
and
self-efficacy judgment.
If
if
low confidence
a necessary condition for learning from
feedback, then effectively using feedback to gauge one's performance
may
Overconfidence
in
Self-Efficacy
24
require a malleable sense of self-efficacy. While researchers have argued the
benefits of consistent
and unchanging self-perceptions
(e.g.,
Wood &
Bandura,
1989), self-efficacy research has given less attention to the potential for rigid self-
perceptions to circumvent learning
It
may be necessary
(cf.
Einhorn, 1980; Einhorn
to reinterpret
some previous
&
Hogarth, 1978).
self-efficacy research
considering the issue of overconfidence. For example, Eden and Kinnar (1991)
used a special
training
program
to increase the self-efficacy of Israeli
Force (IDF) draftees. The manipulation was successful
in
Defense
increasing the rate of
volunteering for IDF special forces. However, the only dependent variable
study
was a commitment
forces).
to
perform a future act
(i.e.,
serving
Without objective measures of draftee performance,
in
it
in
the
the special
is
impossible to
say whether the manipulation increased the draftees' overconfidence or induced
appropriate confidence. However, the robustness of overly positive selfevaluations prior to having first-hand task experience suggests the former
explanation
is
more
likely
than the
latter.
This research also provides evidence that expectations of benefits from an
innovation can affect self-efficacy, independent of any substantive benefits from
the innovation (King, 1974; Stone, 1992).
efficacy
changed as the
result of
hardware and software. The
In
the experiment, participants' self-
whether they expected help from computer
results therefore provide support for the
argument
that the implementation of innovations offers the opportunity to simultaneously
change
self-efficacy (Eden, 1990).
However, participants with mildly negative
expectations performed best, which suggests that expectations of benefits from an
innovation that are either unrealistically high or low
least in
complex cognitive
tasks.
Similar results
may
appear
hurt task performance, at
in field
research.
For
example, Ginzberg (1981) found that computer system users with extreme
expectations
(i.e.,
very high or very low) of system performance used a bank
Overconfidence
computer system less and were less
realistic
25
Self-Efficacy
This suggests that
satisfied with the system.
may be most
expectations of organizational innovations
in
effective,
which
contrasts with prior suggestions (e.g., Eden, 1990; King, 1974) that positive
expectations of organizational innovations produce the greatest productivity
Given the increasing reliance on information technology
gains.
in
work
settings,
the relationship between self-efficacy and organizational innovations intended to
improve productivity and decision making
on
processes (Stone,
self-referent
While
asymmetric
self-efficacy
for positive
perform well did
1
an important topic
is
those
to
in initial
judgments affected performance, these effects were
and negative expectations. Participants who expected
not, but
believed they did.
self-efficacy
who expect
to
do
research
992).
Participants
performance outperformed those who expected
biases
for future
to
who expected
to
"average"
perform badly. Overconfidence
judgments may therefore induce complacency among
well, but
motivate those
who
believe they must work hard
achieve superior performance. Similar results have occurred
in
response
to
manipulations of the type of feedback given to decision makers. Bandura and
Jourden (1991) gave some participants engaged
feedback
that led
them
participants received
to believe they
feedback
task through hard work.
that led
Participants
had
easily
them
who
in
a decision making task
mastered the
to believe they
task.
Other
had mastered the
believed they had easily mastered the
task set lower goals for themselves and performed worse than participants
believed they had worked hard to achieve mastery.
The authors argue
who
that
conditions which create "complacent self-assurance" provide few incentives for
exerting the effort necessary for attaining high levels of achievement.
complementary explanation
condition participants
in this
for the superior
study
is
A
performance of the mildly negative
that they perceived the task to
be more
challenging and therefore expenenced "flow" conditions (Csikzentmihaiyi, 1990;
Overconfidence
Csikzentmihaiyi
& LeFevre, 1989)
in
which both
skills
If,
between perceived
skills
expectations
26
and task demands were
and challenges often produces the highest
involvement and performance, the presence of overconfidence
that
Self-Efficacy
as Csikzentmihaiyi and colleagues argue, a balance
perceived as high.
may mean
in
levels of
in self-efficacy
moderating overly positive self-evaluations with mildly negative
may
actually improve performance under certain conditions.
therefore also be true that, contrary to suggestions
in prior
may
research, mildly
negative self-appraisals enhance performance relative to overly positive
appraisals under certain conditions.
It
self-
Overconfidence
in
Self-Efficacy
27
References
Bandura, A. (1977). Self-efficacy: Toward a unifying theory of behavioral
change. Psychological Review, 84, 191-215.
Bandura, A. (1982). Self-efficacy mechanism
Psychologist, 37, 122-147.
Bandura, A.
(1
theory,
Bandura,
A.,
A.,
human agency. American
Social foundations of thought
986).
Englewood
& Adams,
Cliffs:
Adams,
and action: A
social cognitive
Prentice-Hall.
(1977). Analysis of self-efficacy theory of
N. E.
behavioral change.
Bandura,
in
Cognitive Therapy
N. E., Hardy, A. B.,
generality of self-efficacy theory.
and Research,
1,
287-308.
&
Howells, G. N. (1980). Tests of the
Cognitive Therapy and Research, 4, 39-
66.
Bandura, A., & Cervone, D. (1983). Self-evaluative and self-efficacy mechanisms
governing the motivational effects of goal systems. Journal of Personality
and Social Psychology, 45:5, 1017-1028.
Bandura,
A.,
& Cervone,
influences
in
D.
Differential
(1986).
cognitive motivation.
Decision Processes, 38, 92-1
1
engagement
of self-reactive
Organizational Behavior
and Human
3.
Bandura, A., & Jourden, F. J. (1991). Self-regulatory mechanisms governing the
impact of social comparison on complex decision making. Journal of
Personality and Social Psychology 60:6, 941-951
Bandura,
A.,
& Schunk,
D. H.
(1981).
Cultivating competence, self-efficacy,
and
through proximal self-motivation. Journal of Personality
Social Psychology, 41:3, 586-598.
intrinsic interest
and
Bandura, A., & Wood, R. E. (1989). Effect of perceived controllability and
performance standards on self-regulation of complex decision making.
Journal of Personality and Social Psychology, 56, 805-814.
Baron, R. M.,
& Kenney,
D. A.
(1986).
The moderator-mediator
variable
distinction in social psychological research: conceptual, strategic,
statistical
considerations Journal of Personality
and
and
Social Psychology.
51:6,1173-1182.
Cervone, D.
(in press).
motivation,
New York:
The
role of self-referent cognitions in goal setting,
and performance,
Ablex.
in
M. Robinowitz (Ed.), Applied Cognition,
.
Overconfidence
Cervone,
D., Jiwani, N.,
& Wood,
R.
(1991). Goal setting
and the
in
Self-Efficacy
28
differential
influence of self-regulatory processes on complex decision-making
performance Journal of Personality and Social Psychology, 61 257-26
,
Cervone,
D.,
&
W.
Palmer, B.
Cognitive Therapy
self-efficacy beliefs.
Cervone,
D.,
& Peake,
(1990). Anchoring biases
P. K.
and the perseverance of
and Research,
14, 401-416.
and action: The
on self-efficacy judgments and behavior.
(1986). Anchoring, efficacy,
influence of judgmental heuristics
Journal of Personality and Social Psychology, 50, 492-501
Cervone,
D.,
& Wood,
R.
(1992). Goals, feedback,
on a complex decision
of self-regulatory processes
University of
Clarke, F. R.
Illinois at
(1960).
differential influence
task.
Working paper,
Chicago.
Confidence
confusion matrices
and the
second-choice responses, and
tests. Journal of The Acoustical Society
ratings,
in intelligibility
of America, 32, 35-46.
Solomon, D., & Kelleway, R. (1955). Expectancy statements and
decision times as functions of objective probabilities and reinforcement
Crandall, V.
J.,
values. Journal of Personality, 24, 192-203.
Csikzentmihaiyi, M. (1990). Flow: The psychology of optimal experience.
York: Harper and Row.
&
Csikzentmihaiyi, M.,
LeFevre,
Eden, D. (1990). Pygmalion
D.,
&
(1989). Optimal experience
in
work and
Journal of Personality and Social Psychology, 56, 81 5-822.
leisure.
Eden,
J.
New
Kinnar,
J.
in
management. Lexington: Lexington Books.
(1991). Modeling galatea: Boosting self-efficacy to
increase volunteering. Journal of Applied Psychology, 76, 770-780.
Learning from experience and suboptimal rules in decision
making. In T. Wallsten (Ed.), Cognitive Processes in Choice and Decision
Behavior. Hillsdale, NJ: Lawrence Erblaum.
Einhorn, H.
J.
Einhorn, H.
J.,
(1980).
&
Hogarth, R. M. (1978). Confidence
the illusion of
Festinger,
L.
validity.
(1942).
judgment: persistence of
Psychological Review, 35:3, 395-416.
A theoretical
in
interpretation of shifts in level of aspiration.
Psychological Review, 87, 215-251.
Festinger,
L.
(1954).
A theory
Relations, 7, 117-140.
of social
comparison processes.
Human
.
Overconfidence
Fischhoff, B.,
&
Slovic, P.
(1980).
Nickerson
Hillsdale: Erlbaum.
judgment.
In R.
Fischhoff, B., Slovic, P.,
&
A little
(Ed.),
learning:
Attention
Lichtenstein, S.
Confidence
in
in
Knowing with
29
multicue
and Performance
(1977).
Self-Efficacy
VIII.
certainty:
The
appropriateness of extreme confidence. Journal of Experimental
Psychology: Human Perception and Performance, 3, 552-564.
Frank, J.D. (1953).
Some
psychological determinants of the level of aspiration.
American Journal of Psychology, 47, 285-293.
Gigerenzer, G., Hoffrage, U., & Kleinbolting, H. (1991). Probabilistic mental
models: A brunswikian theory of confidence. Psychological Review, 98:4,
506-528.
Ginzberg, M.J. (1981). Early diagnosis of MIS implementation failure: promising
results and unanswered questions. Management Science, 27 A, 459-478.
Gist,
M. E., & Mitchell, T. R. (1992). Self-efficacy: A theoretical analysis of its
determinants and malleability. The Academy of Management Review,
17:2, 183-211.
Goethals, G. R., & Darley, J. M. (1977). Social comparison theory: Attributional
approach. In J. M. Suls & R. L. Miller (Eds.), Social comparison
processes: theoretical and empirical perspectives Washington, DC:
Hemisphere.
Green, S.
K.,
&
Gross, A.
Personality
and
E.
(1979).
Self-serving biases
Social Psychology Bulletin,
5,
in implicit
evaluations.
214-217.
Greenwald, A. G. (1980). The totalitarian ego: Fabrication and revision of
personal history. American Psychologist, 35, 603-618.
Hays,
W. L
(1981).
Smith, N. D.,
Hill, T.,
Statistics,
New York:
& Mann, M.
predicting the decision to
Holt, Rinehart,
and Winston.
Role of efficacy expectations in
use advanced technologies. Journal of Applied
G. (1987).
Psychology, 72, 307-314.
Hoch,
S. J.
(1985).
Counterfactual reasoning and accuracy
in
predicting
personal events. Journal of Experimental Psychology. 11,71 9-731
Hogarth, R. (1983). Judgment and Choice.
InA/in, F.
W.
126.
(1944).
The
New York: John
Wiley.
realism of expectations. Psychological Review, 51, 120-
Overconfidence
in
30
Self-Efficacy
W., Schkade, D. A., & Bettman, J. R. (1989).
Monitoring information processing and decisions: The mouselab system.
Working paper, University of Pennsylvania.
Johnson,
E. J.,
Payne,
J.
Jourden, F. (1992). The influence of feedback on self-regulatory mechanisms:
glass half full or half empty. Working paper, University of Illinois at
A
Urbana-Champaign.
Keren, G. (1991). Calibration and probability judgments: Conceptual and
methodological issues. Acta Psyciiologica, 77, 217-273.
Expectation effects
Science Quarterly, 19, 221-230.
King, A. S.
Klayman,
(1974).
J.
in
organizational change. Administrative
(1983). Analysis of predecisional information search patterns.
In
Humphreys, O. Svenson, & A. Vari (Eds.), Analysis and Aiding
Decision Processes, Amsterdam: North Holland.
P.C.
&
Ha, Y. (1987). Confirmation, disconfirmation, and information
hypothesis testing. Psychological Review, 94:2, 21 1-228.
Klayman,
J.,
&
Koriat, A., Lichtenstein, S.,
Fischhoff, B.
Journal of Experimental Psychology:
(1980).
Reasons
Human
Learning and Memory,
in
for confidence.
6,
107-
118.
Kuiper, N. A.,
&
reference
Depressed and nondepressed content
mild depression. Journal Of Personality, 50, 67-79.
Derry, P. A.
in
(1982).
& MacDonald,
M. R. (1982). Self and other perception
depressives. Social Cognition, 1 233-239.
Kuiper, N. A.,
in
self-
mild
,
MacDonald, M. R., & Shaw, B. F. (1985). Selfschema processing of depressed and nondepressed content: The effects of
vulnerability on depression. Social Cognition, 3, 77-93.
Kuiper, N. A., dinger,
L. J.,
Langer, E.
The
J.
(1975).
illusion of control.
Journal of Personality and Social
Psychology, 32, 311-328.
Lichtenstein, S., Fischhoff, B.,
probabilities:
The
Phillips, L. D.
(1982). Calibration of
state of the art to 1980.
Tversky, A. (eds).
Hillsdale:
&
In:
Kahneman,
Judgment under uncertainty:
D., Slovic, P.
Heuristics
and biases.
ErIbaum.
Locke, E. A. (1991). Introduction to special issue on cognitive self-regulation.
Organizational Behavior and Human Decision Processes, 50, 151-153.
&
Overconfidence
Locke, E. A., & Latham, G. P. (1990). A theory of goal setting
performance. Englewood Cliffs: Prentice-Hall.
Murphy, A.
&
H.,
Winkler, R.
(1977).
L.
Nickerson, R.
&
S.,
31
Self-Efficacy
and task
Can weather forecasters formulate
reliable probability forecasts of precipitation
Weatlier Digest,
in
and termparture? National
2, 2-9.
McGoldrick, C.
C,
performance on a judgemental
Jr.
Confidence ratings and level of
Perceptual Motor Skills, 20, 31 1-316.
(1965).
task..
Paese,
W., & Sniezek, J. A. (1991). Influences on the appropriateness of
confidence in judgment: practice, effort, information, and decision-making.
Organization Behavior and Human Decision Processes, 48, 100-130.
Payne,
Task complexity and contingent processing in decision
making: An information search and protocol analysis. Organizational
Behavior and IHuman Performance, 16, 366-387.
Payne,
P.
J.
W.
J.
W., Braunstein, M.
(1976).
behavior:
An
Behavior and
Peake,
P. K.,
L.,
&
Carroll,
J.
S.
(1978).
Exploring predecisional
approach to decision research. Organizational
Performance, 22, 17-44.
alternative
Human
& Cervone,
Primary effects
in
D.
(1989).
Sequence anchoring and
the consideration of possibilities.
self-efficacy:
Social Cognition,
7,
31-
50.
Pitz,
G. F. (1974). Subjective probability distributions for imperfectly
quantities.
In L.
W. Gregg
(Ed.),
Knowledge and
cognition.
known
New York:
Wiley.
Sarason,
C.
Sarason,
I.
G. (1975). Anxiety and self-preoccupation.
Spielberger (Eds.), Stress and Anxiety,
I.
G.,
Sarason,
interference,
5,
B. R.,
&
Pierce, G. R.
2,
In
I.
G. Sarason and D.
27-44.
(1990). Anxiety, cognitive
and performance. Journal of Social Behavior and Personality,
1-18.
Schunk, D.
(1983). Ability versus effort attributional feedback: Differential
H.
effects
on
self-efficacy
and achievement. Journal of Educational
Psychology, 75, 848-856.
Schunk, D. H. (1984). Sequential attributional feedback and children's
achievement behavior. Journal of Educational Psychology, 76, 1 1 59-1 1 69.
Sniezek,
J. A.,
&
Buckley, T. (1991). Confidence
depends on
aggregation. Journal of Behavioral Decision Making,
4,
level of
263-272.
Overconfidence
in
Self-Efficacy
32
Stone, D. (1992). Qualitative information technology as self-fulfilling prophecy. In
M. Silver (Ed.), DSS'92 Transactions: Twelfth International Conference on
Decision Support Systems, Providence: Institute of Management Science.
Stone, D., & Schkade, D. A. (1991). Numeric and linguistic information
representation in multiattribute choice. Organizational Behavior and
Human Decision Processes, 49, 42-59.
Suls,
J.
M.,
&
Miller, R. L.
(1977).
Social comparison processes: Theoretical
and
empirical perspectives. Washington, DC: Hemisphere.
Swets,
Tanner, W. P., Jr., & Birdsall, T. (1961). Decision processes
perception. Psychological Review, 68, 301-340.
J. A.,
in
&
Sniezek, J. A. (1991). Judgment processes in motivation:
Anchoring and adjustment effects on judgment and behavior.
Switzer, F. S.,
Organizational Behavior
Taylor, S. E.,
& Brown,
J.
and Human Decision Processes,
(1988).
D.
Illusion
and well-being: A
49, 208-229.
social
psychological perspective on mental health. Psychological Bulletin, 103,
193-210.
Tomassini,
L. A.,
Solomon,
I.,
Romney, M.
B.,
&
Drogstad,
Calibration of auditors' probabilistic judgments:
Organizational Behavior
J. L.
some
and Human Performance,
(1982).
empirical evidence.
30, 391-406.
Trafimow, D., & Sniezek, J. A. (in press). Perceived expertise and its effect on
confidence. Organizational Behavior and Human Decision Processes.
& Bandura,
Wood,
Impact of conceptions of ability on selfregulatory mechanisms and complex decision-making. Journal of
Personality and Social Psychology, 56, 407-415.
Wood,
Locke, E. A. (1987). The relation of self-efficacy and grade goals
to academic performance. Educational and Psychological Measurement,
R. E.,
R. E.,
&
47, 1013-1024.
A.
(1989).
Overconfidence
in
Self-Efficacy
33
Author Notes
Sincere thanks
to the
Peat Marwick Research Foundation and the Office of
Accounting Research (Department of Accountancy, University of
financial support
Illinois) for
Eden
and
to the Office for Information
assistance with the experiments.
Management
In addition,
Illinois
correspondence
to
for helpful
Dan
Ira
comments on previous
Stone, University of
and Information Systems, 1206
S. Sixth St.,
[email protected].
Thanks also
to
Dov
my
(especially Ciif Brown, Daniel Cervone,
Forest Jourden, Kathryn Kadous, Janet Sniezek, and
anonymous reviewers
(University of
conversations with
(Tel Aviv University) helped in planning the research.
colleagues at the University of
Illinois) for
Illinois,
Solomon) and
drafts.
to
two
Send
Department of Accountancy
Champaign,
IL
61820, email:
Overconfidence
in
Self-Efficacy
34
Footnotes
There
''
efficacy (Gist
is
&
considerable variability
Mitchell, 1992).
measures should be used
efficacy (e.g.,
researchers argue that multiple, sequential
measure both the strength and magnitude
measures may be redundant
introduce systematic biases (Peake
measures
single, well-defined
responses
to a set of Likert-type scales to
1977;
Smith,
A cash
2
pilot
& Mann,
prize
was
(e.g..
& Cervone,
have used
Hill,
the methods used to assess self-
Locke & Latham, 1990). However, there
multiple, sequential
may
to
Some
in
(e.g.,
some evidence
Wood &
that
Locke, 1987) and
1989). Other researchers
Cervone & Peake, 1986) or
assess
Bandura,
self-efficacy (e.g.,
1987; Schunk, 1983, 1984).
offered to increase the rate of participation.
In
both the
study and experiment, participants were (correctly) told that cash prizes
would be randomly distributed and
chances
that task
performance would not
search pattern (F =
p=
affect their
of winning.
3 There were no significant differences due
2.2,
is
of self-
.
1 1
)
1 .4,
p=
(See Payne,
1
.26) or the
976
for
to expectation condition in the
percentage of information searched (F =
a description of these measures).
Overconfidence
Table
in
Self-Efficacy
35
1
Results of Experiment
Deviations by Expectation Condition
Means and Standard
iiiun
Positive
(n=47)
Mildly
Strongly
Negative
(n=45)
Negative
(n=47)
F(2.136)
56.3b
31.3c
115.0
<.01
17.2
<.01
3.7
.03
Manipulation Checks
1
.
Expected Accuracy
Percentile
2.
4.
Rank
(9.0)
Expected help from
hardware and software
Performance.
3.
88.83
Effort,
5.63
(1.3)
to
52.7
(30.1)
Choice
111.5
(39.5)
5.
Variability in
Time
6.0
Spent Per Alternative
6. Variability in
3.8b
(24.7)
4.0b
(1.7)
(1.8)
56.43
41. lb
and Information Search
Actual Accuracy
Percentile Rank
Time
(18.3)
(2.8)
Time
5.8
Spent Per Attribute
(28.2)
126.13
(66.3)
7.23
(4.6)
6.33
(26.7)
87.9b
6.8
<
6.8
< .01
3.2
.05
.01
(39.8)
4.6b
(2.3)
4.7b
(2.5)
(3.9)
(2.3)
79.53
54.5b
32.3c
Post-ExDerimental Measures
7.
Perceived Choice Accuracy
Percentile
8.
Rank
Perceived Help from
Hardware and Software
Standard deviations are shown
3' b-
(13.5)
c Entries within
4.83
(1.5)
in
(17.2)
2.8b
(1.9)
69.1
<.01
15.5
<.01
(25.7)
3.2b
(2.0)
parentheses.
a row with different letters are significantly different according to a Tukey
Honestly Significant Difference (HSD) test (p < .05).
Overconfidence
in
Self-Efficacy
Table 2
Results of Experiment
Number of Participants by Actual Expectations and Expectation Condition
(Positive and Mildly Negative Conditions Only)
Expectation Condition
Positive
Mildly Negative
Total
Expectation > 69.6
45
11
56
Expectation < 69.6
2
34
36
47
45
92
Actual Expectations:
Total
36
CO
o
o
a
o
y
o
o
ri-
(D
CO
CO
91
(D
n
ve
o
CO
o
CD
\
N
91
(D
o
o
3
O
3
»
^
dvC
d-
(9
III
III
III
n
W
o
n
o
H>
CO
T3
l->
03
o
H-
C
rr
i-l
q:
h-
3
II
O
s
a.
III
III
III
III
W
«
o