Content uploaded by Eric J Johnson
Author content
All content in this area was uploaded by Eric J Johnson
Content may be subject to copyright.
Journal
of
Personality
and
Social Psychology
1983,
Vol.
45, No. 1,
20-31Copyright
1983
by the
American Psychological
Association,
Inc.
Affect,
Generalization,
and the
Perception
of
Risk
Eric
J.
Johnson
Carnegie-Mellon
University
Amos
Tversky
Stanford University
Experimental manipulations
of
affect
induced
by a
brief newspaper
report
of a
tragic event produced
a
pervasive increase
in
subjects'
estimates
of the
frequency
of
many risks
and
other undesirable
events.
Contrary
to
expectation,
the
effect
was
independent
of the
similarity between
the
report
arid
the
estimated
risk.
An
account
of a
fatal
stabbing
did not
increase
the
frequency
estimate
of a
closely
related risk, homicide, more than
the
estimates
of
unrelated risks such
as
natural
hazards.
An
account
of a
happy event that
created
positive
affect
produced
a
comparable global
decrease
in
judged frequency
of
risks.
As
a
society,
we
have never been more con-
cerned
with
the
assessment,
the
management,
and the
regulation
of
risk. Because public
reaction
to
hazards
from
pesticides, nuclear
power,
or
food
additives appears
to
influence
the
regulation
and
management
of
these
tech-
nologies,
it is
important
to
understand
how
the lay
person perceives
and
evaluates risks.
This
is
particularly true
for
hazards such
as
terrorism, nuclear power,
or
genetic engi-
neering
for
which
the
available statistical data
are
very
limited
and
where
the
assessments
of
the
risks
are
based
on
subjective
and in-
tuitive
judgments.
Indeed,
psychologists
and
other researchers
have
shown increasing interest
in the
manner
in
which people perceive
and
estimate
the
severity
of
various risks. Lichtenstein, Slovic,
Fischhoff,
Layman,
and
Combs
(1978)
asked
lay
people
to
estimate
the
number
of
deaths
per
year that
are due to
various hazards.
These investigators
found
that
the
subjective
estimates were consistent across several
methods
of
elicitation
and
correlated reason-
ably
well
(median
r = .7)
with actuarial
es-
timates
of
frequency.
A
comparison
of ob-
jective
and
subjective estimates revealed
two
This
work
has
been
supported
in
part
by a
National
Science Foundation
postdoctoral
fellowship
to the first
author
and
Office
of
Naval Research
Grant
NR
197-058
to
Stanford University.
We
thank
Baruch
Fischhoff,
Don
MacGregor,
and
Paul Slovic
for
their
useful
comments
and
their generous help
in
conducting Experiments
1
and
2.
Requests
for
reprints should
be
sent
to
Eric
J.
John-
son, Graduate School
of
Industrial
Administration,
Car-
negie-Mellon University, Pittsburgh, Pennsylvania
15213.
biases.
The
primary bias
refers
to the
com-
mon
tendency
to
overestimate infrequent
causes
of
death while underestimating more
frequent
causes. Thus, people estimated that
a
relatively rare cause
of
fatality,
polio,
oc-
curred
about
10
times more frequently than
it
actually does
and
underestimated
the oc-
currence
of a
common cause, lung cancer,
by
about
a
factor
of
10.
The
secondary
bias
re-
fers
to the
observation that "overestimated
causes of
death were generally dramatic
and
sensational whereas underestimated causes
tended
to be
unspectacular events, which
claim
one
victim
at a
time"
(Slovic, Fisch-
hoff,
&
Lichtenstein,
1982,
p.
467). This bias
can
be
attributed
to the
availability
heuristic
in
which
one
estimates
the
frequency
of a
class
by the
ease with which
its
instances
are
brought
to
mind (Tversky
&
Kahneman,
1973).
For
example, homicide
is
perceived
to
occur more frequently than suicide,
and
twice
as
many people
are
thought
to die
from
fire
than
from
drowning. Actually, this
is not
true. About 6,000 more people
die in
suicides
than
in
homicides,
and
drowning causes
death
about
as
frequently
as fire.
Lichtenstein
et
al.
suggested
that
these errors could
reflect
an
availability
bias
induced
by the
media,
which
report homicide
and
fire
more
fre-
quently
than suicide
and
drowning.
A
con-
tent analysis
of
newspaper reports (Combs
&
Slovic,
1979)
supported this interpretation.
Cognition
and
Affect
in
Risk Perception
One
characteristic that distinguishes judg-
ments about risks
from
other estimates, such
as
letter
frequency,
is
that they seldom occur
20
AFFECT,
GENERALIZATION,
AND THE
PERCEPTION
OF
RISK
21
in an
emotionally neutral context. When
we
witness
an
accident
or
read
a
newspaper
re-
port about
a
natural disaster,
we do not
merely
revise
our
subjective
probabilities;
we
are
also
shaken
and
disturbed.
Our
encoun-
ters with news about risk
and
death com-
monly
generate
fear,
anxiety,
and
worry.
Imagine
that
you
have
just read
a
disturbing
newspaper account
_of,
say,
the
senseless
shooting
of an
innocent passerby. Such
an
account,
and the
emotional response
it
elicits,
may
well
increase your estimate
of the
num-
ber
of
deaths
due to
homicide more than
is
justified
by the
information contained
in the
report.
To
investigate
the
role
of
affect
in
judg-
ments
of
risk
we
constructed brief accounts
of
tragic deaths
of a
single person
from
a
spe-
cific
cause. Like many newspaper
stories,
our
accounts described details
of the
tragic inci-
dent
but
gave
no
information about
its
prev-
alence. Thus
the
stories were designed
to
gen-
erate
affect
while conveying minimal data
about
the
frequency
of the
relevant hazard
in
the
general population.
The
effect
of
these
stories
on the
perceived
frequency
of
deaths
caused
by
various risks
was
studied
by
com-
paring estimates made
by two
groups:
one
who
had
read
the
stories
and
another
who
had
not.
We
consider
four
possible
effects
of the
experimental treatment, ordered
by the
range
of
their impact:
1.
The
stories
may
have
no
effect
on es-
timates
of
fatalities. This appears
to be a
nor-
matively
appropriate response because
the
stories, which describe
the
death
of a
single
person,
do not
justify
significant changes
in
frequency
estimates.
2. The
stories
may
produce
a
local
effect.
That
is, a
story about homicide
may
increase
the
estimated
frequency
of
homicide,
and so
forth.
This could
be
caused
by
several mech-
anisms.
The
incident described
in the
story
could serve
as a
retrieval cue, making similar
instances more available, thereby increasing
the
judged
frequency
of the
relevant
risk. The
subject
in the
study, like
the
reader
of a
news-
paper, could also make
the
inference
that
the
selection
of a
story about
a
particular cause
of
death indicates that this
risk is
especially
serious
or
prevalent. Either mechanism
re-
sults
in a
local
increase,'that
is, a
higher fre-
quency
estimate
for the
cause
of
death men-
tioned
in the
story.
3. The
effect
of the
story could also gen-
eralize
to
other
risks
according
to
their sim-
ilarity
to the risk
described
in the
story.
Hence,
a
story about
a
leukemia victim,
for
example,
will
be
expected
to
have
(a) a
con-
siderable
effect
on the
estimated
frequency
of
other cancers,
(b) a
weaker
effect
on the
estimated
frequency
of
other diseases,
and
(c)
little
or no
effect
on the
estimated fre-
quency
of
unrelated
risks
(e.g., tornado
or
plane crashes). This hypothesis
(which
ap-
peared most probable
to us) is
suggested
by
the
classical notion that
the
gradient
of
gen-
eralization
is
determined
by the
similarity
between
the
critical stimulus
and the
target.
4. The
story could produce
a
global
effect;
that
is, an
increase
in the
judged
frequency
of
both related
and
unrelated risks. There
is
evidence
in the
social psychological
literature
that
changes
in
affect
could
influence
a
wide
range
of
unrelated behaviors
and
judgments
such
as the
frequency
of
helping
(Isen
&
Levin,
1972)
and the
evaluations
of the
per-
formance
of
household appliances (Isen,
Shalker,
Clark,
&
Karp,
1978).
(For
a
review
of
some
of the
relevant literature,
see
Clark
and
Isen,
1982).
Thus,
we
consider
four
possible
effects
of
the
exposure
to an
affect-arousing
story:
(a)
no
increase
in
fatality
estimates;
(b) a
local
increase, limited
to the
target
risk; (c) a
gen-
eralization gradient;
and (d) a
global increase.
The
four
possible
effects
are not
necessarily
incompatible; both local
and
global increases
could
be
found
in the
same
set of
data.
The Set of
Risks
To
generate
a set of risks for the
present
studies,
we
asked
a
group
of 68
undergrad-
uates
from
Stanford
University
to
list
the
major
risks,
including diseases, accidents,
and
disasters, that
are
important causes
of
death
in the
United States.
Of the 36 risks
listed
by the
subjects,
we
selected
a set of 18
risks
that
were either
the
most
frequent
causes
of
death
or
were listed most frequently
by
the
subjects.
To
ascertain
the
perceived structure
of
these
risks, we
presented
all
pairs
of
18
risks
to
110
subjects recruited through
an
adver-
22
ERIC
J.
JOHNSON
AND
AMOS TVERSKY
tisement
in the
University
of
Oregon student
newspaper. They were asked
to
rate
the
sim-
ilarity
between risks
on a
scale
from
1
(low
similarity)
to 10
(high similarity). Three
groups
of
subjects each rated
one
third
of the
total
of 153
pairs.
The
average similarities
between
risks were analyzed using
the
ADD-
TREE
algorithm (Sattath
&
Tversky, 1977).
The
resulting tree representation
of the
prox-
imity
among
the 18
risks
is
shown
in
Fig-
ure
1.
In
an
ADDTREE
(or
additive tree) repre-
sentation,
the
objects (risks) appear
as the
terminal nodes
of the
tree,
and the
distance
between objects
is the