Perception of Risk

Abstract

Studies of risk perception examine the judgements people make when they are asked to characterize and evaluate hazardous activities
and technologies. This research aims to aid risk analysis and policy-making by providing a basis for understanding and anticipating
public responses to hazards and improving the communication of risk information among lay people, technical experts, and decision-makers.
This work assumes that those who promote and regulate health and safety need to understand how people think about and respond
to risk. Without such understanding, well-intended policies may be ineffective.

Full text

as
potent
as
carcinogens
such
as
symphytine.
If
the
other
pyrrolizidine
alkaloids
in
comfrey
were
as
potent
carcinogens
as
symphytine,
the
possible
hazard
of
a
dailv
cup
of
tea
would
be
HERP
=
0.6%
and
that
of
a
daily
nine
tablets
would
be
HERP=
7.3%.
104.
Agarcus
bisporfs
is
the
most
commonly
eaten
mushroom
in
the
United
States
with
an
estimated
annual
consumption
of
340
million
kilograms
in
1984-85.
Mush-
rooms
contain
various
hvdrazine
compounds,
some
of
which
have
been
shown
to
cause
tumors
in
mice.
Raw
mushrooms
fed
over
a
lifetime
to
male
and
female
mice
induced
bone,
forestomach,
liver,
and
lung
tumors
[B.
Toth
and
J.
Erickson,
Cancer
Res.
46,
4007
(1986)].
The
15-g
raw
mushroom
is
given
as
wet
weight.
The
TD50
value
based
on
the
above
report
is
expressed
as
dry
weight
of
mushrooms
so
as
to
be
comparable
to
other
values
for
TD50
in
Table
1;
90%
of
a
mushroom
is
assumed
to
be
water.
A
second
mushroom,
Gyromitra
escuknta,
has
been
similarlv
studied
and
found
to contain
a
mixture
of
carcinogenic
hvdrazines
[B.
Toth,J.
Environ.
Sci.
Health
C2,
51
(1984)].
These
mushrooms
are
eaten
in
considerable
quantities
in
several
countries,
though
less
frequently
in
the
United
States.
105.
Safrole
is
the
main
component
(up
to
90%)
of
oil
of
sassafras,
formerly
used
as
the
main
flavor
ingredient
in
root
beer
[J.
B.
Wilson,
J.
Assoc.
Off
Anal.
Chem.
42,
696
(1959);
A.
Y.
Leung,
Encyclopedia
of
Common
Natural
Ingredients
Used
in
Food,
Drugs
and
Cosmetics
(Wiley,
New
York,
1980)].
In
1960,
safrole
and
safrole-
containing
sassafras
oils
were
banned
from
use
in
foods
in
the
United
States
[Fed.
Regist.
25,
12412
(1960)].
Safrole
is
also
naturally
present
in
the
oils
of
sweet
basil,
cinnamon
leaf,
nutmeg,
and
pepper.
106.
Diet
cola
available
in
a
local
market
contains
7.9
mg
of
sodium
saccharin
per
fluid
ounce.
107.
Metronidatole
is
considered
to
be
the
drug
of
choice
for
trichomonal
and
Gardnerella
infections
[AMA
Division
of
Drugs,
AMA
Drug
Evaluations
(Ameri-
can
Medical
Association,
Chicago,
IL,
ed.
5,
1983),
pp.
1717
and
1802].
108.
Isoniazid
is
used
both
prophylactically
and
as
a
treatment
for
active
tuberculosis.
The
adult
prophylactic
dose
(300
mg
daily)
is
continued
for
1
year
[AMA
Division
of
Drugs,
AMA
Drug
Evaluations
(American
Medical
Association,
Chicago,
IL,
ed.
5,
1983),
pp.
1766-1777].
109.
D
M.
Siegal,
V.
H.
Frankos,
M.
A.
Schneiderman,
Reg.
Toxicol.
Pharmacol.
3,
355
(1983).
110.
Supported
by
NCI
Outstanding
Investigator
Grant
CA39910
to
B.N.A.,
NIEHS
Centc
Grant
ES01896,
and
NIEHS/DOE
Interagency
Agreement
222-YOl-ES-
10066.
We
are
indebted
to
numerous
colleagues
for
criticisms,
particularly
W.
Havender,
R.
Peto,
J.
Cairns,
J.
Miller,
E.
Miler,
D.
B.
Clayson,
J.
McCann,
and
F.
J.
C.
Roe.
Perception
of
Risk
PAUL
SLOVIC
Studies
of
risk
perception
examine
the
judgments
people
make
when
they
are
asked
to
characterize
and
evaluate
hazardous
activities
and
technologies.
This
research
aims
to
aid
risk
analysis
and
policy-making
by
(i)
providing
a
basis
for
understanding
and
anticipating
public
responses
to
hazards
and
(ii)
improving
the
communication
of
risk
information
among
lay
people,
technical
experts,
and
decision-makers.
This
work
assumes
that
those
who
pro-
mote
and
regulate
health
and
safety
need
to
understand
how
people
think
about
and
respond
to
risk.
Without
such
understanding,
well-intended
policies
may
be
inef-
fective.
T
HE
ABILITY
TO
SENSE
AND
AVOID
HARMFUL
ENVIRONMEN-
tal
conditions
is
necessary
for
the
survival
of
all
living
organisms.
Survival
is
also
aided
by
an
ability
to
codify
and
learn
from
past
experience.
Humans
have
an
additional
capability
that
allows
them
to
alter
their
environment
as
well
as
respond
to
it.
This
capacity
both
creates
and
reduces
risk.
In
recent
decades,
the
profound
development
of
chemical
and
nuclear
technologies
has
been
accompanied
by
the
potential
to
cause
catastrophic
and
long-lasting
damage
to
the
earth
and
the
life
forms
that
inhabit
it.
The
mechanisms
underlying
these
complex
technolo-
gies
are
unfamiliar
and
incomprehensible
to
most
citizens.
Their
most
harmful
consequences
are
rare
and
often
delayed,
hence
difficult
to
assess
by
statistical
analysis
and
not
well
suited
to
management
by
trial-and-error
learning.
The
elusive
and
hard
to
manage
qualities
of
today's
hazards
have
forced
the
creation
of
a
new
intellectual
discipline
called
risk
assessment,
designed
to
aid
in
identifying,
characterizing,
and
quantifying
risk
(1).
Whereas
technologically
sophisticated
analysts
employ
risk
assess-
ment
to
evaluate
hazards,
the
majority
of
citizens
rely
on
intuitive
risk
judgments,
typically
called
"risk
perceptions."
For
these
people,
280
experience
with
hazards
tends
to
come
from
the
news
media,
which
rather
thoroughly
document
mishaps
and
threats
occurring
throughout
the
world.
The
dominant
perception
for
most
Ameri-
cans
(and
one
that
contrasts
sharply
with
the
views
of
professional
risk
assessors)
is
that
they
face
more
risk
today
than
in
the
past
and
that
future
risks
will
be
even
greater
than
today's
(2).
Similar
views
appear
to
be
held
by
citizens
of
many
other
industrialized
nations.
These
perceptions
and
the
opposition
to
technology
that
accompa-
nies
them
have
puzzled
and
frustrated
industrialists
and
regulators
and
have
led
numerous
observers to
argue
that
the
American
public's
apparent
pursuit
of
a
"zero-risk
society"
threatens
the
nation's
political
and
economic
stability.
Wildavsky
(3,
p.
32)
commented
as
follows
on
this
state
of
affairs.
How
extraordinary!
The
richest,
longest
lived,
best
protected,
most
resourceful
civilization,
with
the
highest
degree
of
insight
into
its
own
technology,
is
on
its
way
to
becoming
the
most
frightened.
Is
it
our
environment
or
ourselves
that
have
changed?
Would
people
like
us
have
had
this
sort
of
concern
in
the
past?
.
.
.
Today,
there
are
risks
from
numerous
small
dams
far
exceeding
those
from
nuclear
reactors.
Why
is
the
one
feared
and
not
the
other?
Is
it
just
that
we
are
used
to
the
old
or
are
some
of
us
looking
differently
at
essentially
the
same
sorts
of
experience?
During
the
past
decade,
a
small
number
of
researchers
has
been
attempting
to
answer
such
questions
by
examining
the
opinions
that
people
express
when
they
are
asked,
in
a
variety
of
ways,
to
evaluate
hazardous
activities,
substances,
and
technologies.
This
research
has
attempted
to
develop
techniques
for
assessing
the
complex
and
subtle
opinions
that
people
have
about
risk.
With
these
techniques,
researchers
have
sought
to
discover
what
people
mean
when
they
say
that
something
is
(or
is
not)
"risky,"
and
to
determine
what
factors
underlie
those
perceptions.
The
basic
assumption
underlying
these
efforts
is
that
those
wvho
promote
and
regulate
health
and
safety
need
to
understand
the
ways
in
which
people
think
about
and
respond
to
risk.
The
author
is
prcsident
of
Decision
Research,
1201
Oak
Street,
Eugene,
OR
97401,
and
professor
of
psychology
at
the
University
of
Oregon.
SCIENCE,
VOL.
236
on February 1, 2010 www.sciencemag.orgDownloaded from

If
successful,
this
research
should
aid
policy-makers
by
improving
communication
between
them
and
the
public,
by
directing
educa-
tional
efforts,
and
by
predicting
public
responses
to
new
technolo-
gies
(for
example,
genetic
engineering),
events
(for
example,
a
good
safety
record
or
an
accident),
and
new
risk
management
strategies
(for
example,
warning
labels,
regulations,
substitute
products).
Risk
Perception
Research
Important
contributions
to
our
current
understanding
of
risk
perception
have
come
from
geography,
sociology,
political
science,
anthropology,
and
psychology.
Geographical
research
focused
orig-
inally
on
understanding
human
behavior
in
the
face
of
natural
hazards,
but
it
has
since
broadened
to
include
technological
hazards
as
well
(4).
Sociological
(5)
and
anthropological
studies
(6)
have
shown
that
perception
and
acceptance
of
risk
have
their
roots
in
social
and
cultural
factors.
Short
(5)
argues
that
response
to
hazards
is
mediated
by
social
influences
transmitted
by
friends,
family,
fellow
workers,
and
respected
public
officials.
In
many
cases,
risk
percep-
tions
may
form
afterwards,
as
part
of
the
ex
post
facto
rationale
for
one's
own
behavior.
Douglas
and
Wildavsky
(6)
assert
that
people,
acting
within
social
groups,
downplay
certain
risks
and
emphasize
others
as
a
means
of
maintaining
and
controlling
the
group.
Psychological
research
on
risk
perception,
which
shall
be
my
focus,
originated
in
empirical
studies
of
probability
assessment,
utility
assessment,
and
decision-making
processes
(7).
A
major
development
in
this
area
has
been
the
discovery
of
a
set
of
mental
strategies,
or
heuristics,
that
people
employ
in
order
to
make
sense
out
of
an
uncertain
world
(8).
Although
these
rules
are
valid
in
some
circumstances,
in
others
they
lead
to
large
and
persistent
biases,
with
serious
implications
for
risk
assessment.
In
particular,
laboratory
research
on
basic
perceptions
and
cognitions
has
shown
that
difficul-
ties
in
understanding
probabilistic
processes,
biased
media
coverage,
misleading
personal
experiences,
and
the
anxieties
generated
by
life's
gambles
cause
uncertainty
to
be
denied,
risks
to
be
misjudged
(sometimes
overestimated
and
sometimes
underestimated),
and
judgments
of
fact
to
be
held
with
unwarranted
confidence.
Experts'
judgments
appear
to
be
prone
to
many
of
the
same
biases
as
those
of
the
general
public,
particularly
when
experts
are
forced
to
go
beyond
the
limits
of
available
data
and
rely
on
intuition
(8,
9).
Research
further
indicates
that
disagreements
about
risk
should
not
be
expected
to
evaporate
in
the
presence
of
evidence.
Strong
initial
views
are
resistant
to
change
because
they
influence
the
way
that
subsequent
information
is
interpreted.
New
evidence
appears
reliable
and
informative
if
it
is
consistent
with
one's
initial
beliefs;
contrary
evidence
tends
to
be
dismissed
as
unreliable,
erroneous,
or
unrepresentative
(10).
When
people
lack
strong
prior
opinions,
the
opposite
situation
exists-they
are
at
the
mercy
of
the
problem
formulation.
Presenting
the
same
information
about
risk
in
different
ways
(for
example,
mortality
rates
as
opposed
to
survival
rates)
alters
people's
perspectives
and
actions
(11).
The
Psychometric
Paradigm
One
broad
strategy
for
studying
perceived
risk
is
to
develop
a
taxonomy
for
hazards
that
can
be
used
to
understand
and
predict
responses
to
their
risks.
A
taxonomic
scheme
might
explain,
for
example,
people's
extreme
aversion
to
some
hazards,
their
indiffer-
ence
to
others,
and
the
discrepancies
between
these
reactions
and
opinions
of
experts.
The
most
common
approach
to
this
goal
has
employed
the
psychometric
paradigm
(12,
13),
which
uses
psycho-
physical
scaling
and
multivariate
analysis
techniques
to
produce
17
APRIL
I987
quantitative
representations
or
"cognitive
maps"
of
risk
attitudes
and
perceptions.
Within
the
psychometric
paradigm,
people
make
quantitative
judgments
about
the
current
and
desired
riskiness
of
diverse
hazards
and
the
desired
level
of
regulation
of
each.
These
judgments
are
then
related
to
judgments
about
other
properties,
such
as
(i)
the
hazard's
status
on
characteristics
that
have
been
hypothesized
to
account
for
risk
perceptions
and
attitudes
(for
example,
voluntariness,
dread,
knowledge,
controllability),
(ii)
the
benefits
that
each
hazard
provides
to
society,
(iii)
the
number
of
deaths
caused
by
the
hazard
in
an
average
year,
and
(iv)
the
number
of
deaths
caused
by
the
hazard
in
a
disastrous
year.
In
the
rest
of
this
article,
I
shall
briefly
review
some
of
the
results
obtained
from
psychometric
studies
of
risk
perception
and
outline
some
implications
of
these
results
for
risk
communication
and
risk
management.
Revealed
and
Expressed
Preferences
The
original
impetus
for
the
psychometric
paradigm
came
from
the
pioneering
effort
of
Starr
(14)
to
develop
a
method
for
weighing
technological
risks
against
benefits
in
order
to
answer
the
fundamen-
tal
question,
"How
safe
is
safe
enough?"
His
"revealed
preference"
approach
assumed
that,
by
trial
and
error,
society
has
arrived
at
an
"essentially
optimum"
balance
between
the
risks
and
benefits
associ-
ated
with
any
activity.
One
may
therefore
use
historical
or
current
risk
and
benefit
data
to
reveal
pattems
of
"acceptable"
risk-benefit
trade-offs.
Examining
such
data
for
several
industries
and
activities,
Table
1.
Ordering
of
perceived
risk
for
30
activities
and
technologies
(22).
The
ordering
is
based
on
the
geometric
mean
risk
ratings
within
each
group.
Rank
1
represents
the
most
risky
activity
or
technology.
Activity
League
of
College
Active
or
Women
students
club
Experts
technology
Voters
members
Nuclear
power
1
1
8
20
Motor
vehicles
2
53
1
Handguns
3
2
1
4
Smoking
4
3
4
2
Motorcycles
5
6
2
6
Alcoholic
beverages
67
5
3
General
(private)
7
15
11
12
aviation
Police
work
8
8
7
17
Pesticides
9
4
15
8
Surgery
10
11
9
5
Fire
fighting
11
10
6
18
Large
construction
12
14
13
13
Hunting
13
18
10
23
Spray
cans
14
13
23 26
Mountain
climbing
15
22
12
29
Bicycles
16
24
14
15
Commercial
aviation
17
16
18 16
Electric
power
(non-
18
19
19
9
nuclear)
Swimming
19
30
17
10
Contraceptives
20
9
22
11
Skiing
21
25
16
30
X-rays
22
17
24
7
High
school
and
23
26
21
27
college
football
Rairoads
24
23
29
19
Food
preservatives
25
12
28
14
Food
coloring
26
20
30
21
Power
mowers
27
28
25
28
Prescription
antibiotics
28
21
26 24
Home
appliances
29
27
27
22
Vaccinations
30
29 29
25
ARTICLES
28I
on February 1, 2010 www.sciencemag.orgDownloaded from

Starr
concluded
that
(i)
acceptability
of
risk
from
an
activity
is
roughly
proportional
to
the
third
power
of
the
benefits
for
that
activity,
and
(ii)
the
public
will
accept
risks
from
voluntary
activities
(such
as
skiing)
that
are
roughly
1000
times
as
great
as
it
would
tolerate
from
involuntary
hazards
(such
as
food
preservatives)
that
provide
the
same
level
of
benefits.
The
ments
and
deficiencies
of
Starr's
approach
have
been
debated
at
length
(15).
They
will
not
be
elaborated
here,
except
to
note
that
concem
about
the
validity
of
the
many
assumptions
inherent
in
the
revealed
preferences
approach
stimulated
Fischhoff
et
al.
(12)
to
conduct
an
analogous
psychometric
analysis
of
questionnaire
data,
resulting
in
"expressed
preferences."
In
recent
years,
numerous
other
studies
of
expressed
preferences
have
been
carried
out
within
the
psychometric
paradigm
(16-24).
These
studies
have
shown
that
perceived
risk
is
quantifiable
and
predictable.
Psychometric
techniques
seem
well
suited
for
identify-
ing
similarities
and
differences
among
groups
with
regard
to
risk
perceptions
and
attitudes
(Table
1).
They
have
also
shown
that
the
Factor
2
Unknown
risk
Laetrile*
Microwave
ovens
-
Water
Fluoridation0
Saccharin
.
Ni
tri
tesl
*
Hexachlorophene
Water
Chlorination
i
Polyvinyl
Ch'orideg
Coal
Tar
Hairdyes5
Oral
Contraceptivese
*
Diagnostic
X
Rays
Valium
DarvonO*
*
IuD
Antibiotics*
Rubber
Mfg
-
*
Caffeine
*
Aspi
ri
n
Auto
Lead
T
*
Lead
Paint
*
Vaccilnes
Power
Mowers
4
I
' I
'
'
'
'
w
' '
Skateboards
*
Smoking
(Disease)
*
Snowmobiles
*
rrampolines
0
*
Tractors
Alcohol-
Chainsaws*
*
Elevators
Home
Swimming
Pools5
S
Electric
Wir
&
Appl
(Fires)
Downhill
Skiing!
*
Smoking
(Fires)
Rec
Boating
5
Electric
Wir
&
Appl
(Shock)o
Bicycles*
Motorcycles*
Bridges5
*
Fireworks
*
*
DNA
Technology
.
*Electric
Fields
*
DES
.-
*
Nitrogen
Fertilizers
0
SST
-
*Cadmium
Usage
*Mirex
*Trichloroethylene
*2,4.5-T
*
Pesticides
6
*
Asbestos
Insulation
*
PCS's
*
Mercury
*DDT
Satellit
*
Fossil
Fuels
*Coal
Burning
(Pollution)
a
a
a
a
a
a
I
_-
*
Auto
Exhaust
(CO)
*
0-CON
*P
Radioactive
Waste
*
Nuclear
Reactor
Accidents
0
Uranium
Mining
*Nuclear
Weapons
Fallout
te
Crashes
Factor
1
Dread
risk
I
SING
SrIr
p
I
N
G
A
ccIe
,
*
LNG
Storage
&
Transport
Nerve
Gas
Accidents
*
Coal
Mining
(Disease)
5
Large
Dams
*
Skyscraper
Fires
Nuclear
Weapons
(War)*
*
Underwater
Const
*
Sport
Parachutes
*
General
Aviation
S
Coal
Mining
Accidents
*
High
Construction
*
Railroad
Collisions
Alcohol
Accidents
*
Coenm
Aviation
*
Auto
Racing
OAuto
Accidents
-
*p
Handguns
*
Dynamite
Factor
2
CONTROLLABLE
NOT
DREAD
NOT
GLOBAL
CATASTROPHIC
CONSEQUENCES
NOT
FATAL
EQUITABLE
INDIVIDUAL
LOW
RISK
TO
FUTURE
GENERATIONS
EASILY
REDUCED
RISK
DECREASING
VOLUNTARY
NOT
OBSERVABLE
UNKNOWN
TO
THOSE
EXPOSED
EFFECT
DELAYED
NEW
RISK
RISKS
UNKNOWN
TO
SCIENCE
UNCONTROLLABLE
DREAD
GLOBAL
CATASTROPHIC
CONSEQUENCES
FATAL
NOT
EQUITABLE
CATASTROPHIC
HIGN
RISK
TO
FUTURE
GENERATIONS
NOT
EASILY
REDUCED
RISK
INCREASING
INVOLUNTARY
Fig.
1.
Location
of
81
hazards
on
factors
1
and
2
derived
from
the
relationships
among
18
risk characteristics.
Each
factor
is
made
up
of
a
combination
of
characteristics,
as
indicated
by
the
lower
diagram
(25).
Factor
1
.
A
I
I
.
.
.
.
i i
. .
.
.
.
.
. .
.
.
. .
.
. . .
o
.
.
...
. . .
. . .
.
. . .
I
I I
v
i
I
.
.
I
I
I
I
v
v
I
I
.
I
.
9
I
.
A
I
I
I
I
I
I 1
-
a
I
a
A
a
a
a
a
I I
I
I
I
.
I
.
I
.
I
I
a
I
l1
I
Ia
a
.
.
.
a
.
.
.
6
-
-T
282
SCIENCE3
VOL.
236
on February 1, 2010 www.sciencemag.orgDownloaded from

concept
"risk"
means
different
things
to
different
people.
When
experts
judge
risk,
their
responses
correlate
highly
with
technical
estimates
of
annual
fatalities.
Lay
people
can
assess
annual
fatalities
if
they
are
asked
to
(and
produce
estimates
somewhat
like
the
technical
estimates).
However,
their
judgments
of
"risk"
are
related
more
to
other
hazard
characteristics
(for
example,
catastrophic
potential,
threat
to
future
generations)
and,
as
a
result,
tend
to
differ
from
their
own
(and
experts')
estimates
of
annual
fatalities.
Another
consistent
result
from
psychometric
studies
of
expressed
preferences
is
that
people
tend
to
view
current
risk
levels
as
unacceptably
high
for
most
activities.
The
gap
between
perceived
and
desired
risk
levels
suggests
that
people
are
not
satisfied
with
the
way
that
market
and
other
regulatory
mechanisms
have
balanced
risks
and
benefits.
Across
the
domain
of
hazards,
there
seems
to
be
little
systematic
relationship
between
perceptions
of
current
risks
and
benefits.
However,
studies
of
expressed
preferences
do
seem
to
support
Starr's
argument
that
people
are
willing
to
tolerate
higher
risks
from
activities
seen
as
highly
beneficial.
But,
whereas
Starr
concluded
that
voluntariness
of
exposure
was
the
key
mediator
of
risk
acceptance,
expressed
preference
studies
have
shown
that
other
(perceived)
characteristics
such
as
familiarity,
control,
catastrophic
potential,
equity,
and
level
of
knowledge
also
seem
to
influence
the
relation
between
perceived
risk,
perceived
benefit,
and
risk
accept-
ance
(12,
22).
Various
models
have
been
advanced
to
represent
the
relation
between
perceptions,
behavior,
and
these
qualitative
characteristics
of
hazards.
As
we
shall
see,
the
picture
that
emerges
from
this
work
is
both
orderly
and
complex.
Unknown
risk
*
0
0
*
1*0
S
0
.
04
.
*.0
.
4
*0
.0
*
%
0
0
0
0
.0
0
0
C.
0
0
S
0
Dread
risk
0
0
0*
00
00
FIg.
2.
Attitudes
toward
regulation
of
the
hazards
in
Fig.
1.
The
larger
the
point,
the
greater
the
desire
for
strict
regulation
to
reduce
risk
(25).
from
these
characteristics
(25).
Instead,
as
noted
earlier,
experts
appear
to
see
riskiness
as
synonymous
with
expected
annual
mortal-
ity
(26).
As
a
result,
conflicts
over
"risk"
may
result
from
experts
and
lay
people
having
different
definitions
of
the
concept.
The
representation
shown
in
Fig.
1,
while
robust
and
informative,
is
by
no
means
a
universal
cognitive
mapping
of
the
domain
of
hazards.
Other
psychometric
methods
(such
as
multidimensional
scaling
analysis
of
hazard
similarity
judgments),
applied
to
quite
different
sets
of
hazards,
produce
different
spatial
models
(13,
18).
The
utility
of
these
models
for
understanding
and
predicting
behavior
remains
to
be
determined.
Factor-Analytic
Representations
Many
of
the
qualitative
risk
characteristics
are
correlated
with
each
other,
across
a
wide
range
of
hazards.
For
example,
hazards
judged
to
be
"voluntary"
tend
also
to
be
judged
as
"controllable";
hazards
whose
adverse
effects
are
delayed
tend
to
be
seen
as
posing
risks
that
are
not
well
known,
and
so
on.
Investigation
of
these
relations
by
means
of
factor
analysis
has
shown
that
the
broader
domain
of
characteristics
can
be
condensed
to
a
small
set
of
higher
order
characteristics
or
factors.
The
factor
space
presented
in
Fig.
1
has
been
replicated
across
groups
of
lay
people
and
experts
judging
large
and
diverse
sets
of
hazards.
Factor
1,
labeled
"dread
risk,"
is
defined
at
its
high
(right-
hand)
end
by
perceived
lack
of
control,
dread,
catastrophic
poten-
tial,
fatal
consequences,
and
the
inequitable
distribution
of
risks
and
benefits.
Nuclear
weapons
and
nuclear
power
score
highest
on
the
characteristics
that
make
up
this
factor.
Factor
2,
labeled
"unknown
risk,"
is
defined
at
its
high
end
by
hazards
judged
to
be
unobserv-
able,
unknown,
new,
and
delayed
in
their
manifestation
of
harm.
Chemical
technologies
score
particularly
high
on
this
factor.
A
third
factor,
reflecting
the
number
of
people
exposed
to
the
risk,
has
been
obtained
in
several
studies.
Making
the
set
of
hazards
more
or
less
specific
(for
example,
partitioning
nuclear
power
into
radioactive
waste,
uranium
mining,
and
nuclear
reactor
accidents)
has
had
little
effect
on
the
factor
structure
or
its
relation
to
risk
perceptions
(25).
Research
has
shown
that
lay
people's
risk
perceptions
and
atti-
tudes
are
closely
related
to
the
position
of
a
hazard
within
this
type
offactor
space.
Most
important
is
the
horizontal
factor
"dread
risk."
The
higher
a
hazard's
score
on
this
factor
(the
further
to
the
right
it
appears
in
the
space),
the
higher
its
perceived
risk,
the
more
people
want
to
see
its
current
risks
reduced,
and
the
more
they
want
to
see
strict
regulation
employed
to
achieve
the
desired
reduction
in
risk
(Fig.
2).
In
contrast,
experts'
perceptions
of
risk
are
not
dosely
related
to
any
of
the
various
risk
characteristics
or
factors
derived
17
APRIL
1987
Accidents
as
Signals
Risk
analyses
typically
model
the
impacts
of
an
unfortunate
event
(such
as
an
accident,
a
discovery
of
pollution,
sabotage,
product
tampering)
in
terms
of
direct
harm
to
victims-deaths,
injuries,
and
damages.
The
impacts
of
such
events,
however,
sometimes
extend
far
beyond
these
direct
harms
and
may
include
significant
indirect
costs
(both
monetary
and
nonmonetary)
to
the
responsible
govern-
ment
agency
or
private
company
that
far
exceed
direct
costs.
In
some
cases,
all
companies
in
an
industry
are
affected,
regardless
of
which
company
was
responsible
for
the
mishap.
In
extreme
cases,
the
indirect
costs
of
a
mishap
may
extend
past
industry
boundaries,
affecting
companies,
industries,
and
agencies
whose
business
is
minimally
related
to
the
initial
event.
Thus,
an
unfortunate
event
can
be
thought
of
as
analogous
to
a
stone
dropped
in
a
pond.
The
ripples
spread
outward,
encompassing
first
the
directly
affected
victms,
then
the
responsible
company
or
agency,
and,
in
the
extreme,
reaching
other
companies,
agencies,
and
industries.
Some
events
make
only
small
ripples;
others
make
larger
ones.
The
challenge
is
to
discover
characteristics
associated
with
an
event
and
the
way
that
it is
managed
that
can
predict
the
breadth
and
seriousness
of
those
impacts
(Fig.
3).
Early
theories
equated
the
magnitude
of
impact
to
the
number
of
people
killed
or
injured,
or
to
the
amount
of
property
damaged.
However,
the
accident
at
the
Three
Mile
Island
(TMI)
nuclear
reactor
in
1979
provides
a
dramatic
demonstration
that
factors
besides
injury,
death,
and
property
damage
impose
serious
costs.
Despite
the
fact
that
not
a
single
person
died,
and few
if
any
latent
cancer
fatalities
are
expected,
no
other
accident
in
our
history
has
produced
such
costly
societal
impacts.
The
accident
at
TMI
devastated
the
utility
that
owned
and
operated
the
plant.
It
also
imposed
enornous
costs
(27)
on
the
nudear
industry
and
on
society,
through
stricter
regulation
(resulting
in
increased
construction
and
operation
costs),
reduced
ARTICLES
283
on February 1, 2010 www.sciencemag.orgDownloaded from

Other
technologies
Ecl
III~
~
~~~~~~nusr
interpretation
Cmpany
r00
Ecal
of
E
go
~E
VctiMs
g
LSignal
Event
Event
Interpretation
character-
istics
Spread
of
impact
Loss
of
sales
Regulatory
constraints
Litigation
Community
opposition
Investor
flight
Type
of
impact
(company
level)
Fig.
3.
A
model
of
impact
for
unfortunate
events.
operation
of
reactors
worldwide,
greater
public
opposition
to
nuclear
power,
and
reliance
on
more
expensive
energy
sources.
It
may
even
have
led
to
a
more
hostile
view
of
other
complex
technologies,
such
as
chemical
manufacturing
and
genetic
engineer-
ing.
The
point
is
that
traditional
economic
and
risk
analyses
tend
to
neglect
these
higher
order
impacts,
hence
they
greatly
underestimate
the
costs
associated
with
certain
kinds
of
events.
Although
the
TMI
accident
is
extreme,
it
is
by
no
means
unique.
Other
recent
events
resulting
in
enormous
higher
order
impacts
include
the
chemical
manufacturing
accident
at
Bhopal,
India,
the
pollution
of
Love
Canal,
New
York,
and
Times
Beach,
Missouri,
the
disastrous
launch
of
the
space
shuttle
Challenger,
and
the
meltdown
of
the
nuclear
reactor
at
Chernobyl.
Following
these
extreme
events
are
a
myriad
of
mishaps
varying
in
the
breadth
and
size
of
their
impacts.
An
important
concept
that
has
emerged
from
psychometric
research
is
that
the
seriousness
and
higher
order
impacts
of
an
unfortunate
event
are
determined,
in
part,
by
what
that
event
signals
or
portends
(28).
The
informativeness
or
"signal
potential"
of
an
event,
and
thus
its
potential
social
impact,
appears
to
be
systemati-
cally
related
to
the
characteristics
of
the
hazard
and
the
location
of
the
event
within
the
factor
space
described
earlier
(Fig. 4).
An
accident
that
takes
many
lives
may
produce
relatively
little
social
disturbance
(beyond
that
experienced
by
the
victims'
families
and
friends)
if
it
occurs
as
part
of
a
familiar
and
well-understood
system
(such
as
a
train
wreck).
However,
a
small
accident
in
an
unfamiliar
Factor
2
Unknown
risk
Accidents
as
signals
0
0
0
0
0
0
0
0
.
0
0
Fig.
4.
Relation
be-
tween
signal
potential
and
risk
characteriza-
tion
for
30
hazards
in
Fig.
1.
The
larger
the
point,
the
greater
the
degree
to
which
an
ac-
cident
involving
that
hazard
was
judged
to
"serve
as
a
warning
signal
for
society,
providing
new
information
about
the
probability
that
similar
or
even
more
destructive
mishaps
might
occur
within
this
type
of
activity."
Media
attention
and
the
higher
order
costs
of
a
mishap
are
likely
to
be
correlated
with
signal
potential
(28).
284
system
(or
one
perceived
as
poorly
understood),
such
as
a
nuclear
reactor
or
a
recombinant
DNA
laboratory,
may
have
immense
social
consequences
if
it
is
perceived
as
a
harbinger
of
further
and
possibly
catastrophic
mishaps.
The
concept
of
accidents
as
signals
was
eloquently
expressed
in
an
editorial
addressing
the
tragic
accident
at
Bhopal
(29).
What
truly
grips
us
in
these
accounts
is
not
so
much
the
numbers
as
the
spectacle
of
suddenly
vanishing
competence,
of
men
utterly
routed
by
technology,
of
fail-safe
systems
failing
with
a
logic
as
inexorable
as
it
was
once-indeed,
right
up
until
that
very
moment-unforeseeable.
And
the
spectacle
haunts
us
because
it
seems
to
carry
allegorical
import,
like
the
whispery
omen
of
a
hovering
future.
One
implication
of
the
signal
concept
is
that
effort
and
expense
beyond
that
indicated
by
a
cost-benefit
analysis
might
be
warranted
to
reduce
the
possibility
of
"high-signal
accidents."
Unfortunate
events
involving
hazards
in
the
upper
right
quadrant
of
Fig.
1
appear
particularly
likely
to
have
the
potential
to
produce
large
ripples.
As
a
result,
risk
analyses
involving
these
hazards
need
to
be
made
sensitive
to
these
possible
higher
order
impacts.
Doing
so
would
likely
bring
greater
protection
to
potential
victims
as
well
as
to
companies
and
industries.
Analysis
of
Single
Hazard
Domains
Psychometric
analyses
have
also
been
applied
to
judgments
of
diverse
hazard
scenarios
within
a
single
technological
domain,
such
as
railroad
transport
(30)
or
automobiles
(31).
Kraus
(30)
had
people
evaluate
the
riskiness
of
49
railroad
hazard
scenarios
that
varied
with
respect
to
type
of
train,
type
of
cargo,
location
of
the
accident,
and
the
nature
and
cause
of
the
accident
(for
example,
a
high-speed
train
carrying
passengers
through
a
mountain
tunnel
derails
due
to
a
mechanical
system
failure).
The
results
showed
that
these
railroad
hazards
were
highly
differentiated,
much
like
the
hazards
in
Fig.
1.
The
highest
signal
potential
(and
thus
the
highest
potential
for
large
ripple
effects)
was
associated
with
accidents
involving
trains
carrying
hazardous
chemicals.
A
study
by
Slovic,
MacGregor,
and
Kraus
(31)
examined
percep-
tions
of
risk
and
signal
value
for
40
structural
defects
in
automobiles.
Multivariate
analysis
of
these
defects,
rated
in
terms
of
various
characteristics
of
risk,
produced
a
two-factor
space.
As
in
earlier
studies
with
diverse
hazards,
the
position
of
a
defect
in
this
space
predicted
judgments
of
riskiness
and
signal
value
quite
well.
One
defect
stood
out
much
as
nuclear
hazards
do
in
Fig.
1.
It
was
a
fuel
tank
rupture
upon
impact,
creating
the
possibility
of
fire
and
burn
injuries.
This,
of
course,
is
similar
to
the
notorious
design
problem
that
plagued
Ford
Pinto
and
that
Ford
allegedly
declined
to
correct
because
a
cost-benefit
analysis
indicated
that
the
correction
costs
greatly
exceeded
the
expected
benefits
from
increased
safety
(32).
Had
Ford
done
a
psychometric
study,
the
analysis
might
have
highlighted
this
particular
defect
as
one
whose
seriousness
and
higher
order
costs
(lawsuits,
damaged
company
reputation)
were
likely
to
be
greatly
underestimated
by
cost-benefit
analysis.
Forecasting
Public
Acceptance
Results
from
studies
of
the
perception
of
risk
have
been
used
to
explain
and
forecast
acceptance
and
opposition
for
specific
technolo-
gies
(33).
Nuclear
power
has
been
a
frequent
topic
of
such
analyses
because
of
the
dramatic
opposition
it
has
engendered
in
the
face
of
experts'
assurances
of
its
safety.
Research
shows
that
pepple
judge
the
benefits
from
nuclear
power
to
be
quite
small
and
the
risks
to
be
unacceptably
great.
Nuclear
power
risks
occupy
extreme
positions
in
SCIENCE,
VOL.
236
a
on February 1, 2010 www.sciencemag.orgDownloaded from

psychometric
factor
spaces,
reflecting
people's
views
that
these
risks
are
unknown,
dread,
uncontrollable,
inequitable,
catastrophic,
and
likely
to
affect
future
generations
(Fig.
1).
Opponents
of
nuclear
power
recognize
that
few
people
have
died
thus
far
as
a
result
of
this
technology.
However,
long
before
Chernobyl,
they
expressed
great
concern
over
the
potential
for
catastrophic
accidents.
These
public
perceptions
have
evoked
harsh
reactions
from
experts.
One
noted
psychiatrist
wrote
that
"the
irrational
fear
of
nuclear
plants
is
based
on
a
mistaken
assessment
of
the
risks"
(34,
p.
8).
A
nuclear
physicist
and
leading
advocate
of
nuclear
power
contended
that
"
.
.
.
the
public
has
been
driven
insane
over
fear
of
radiation
[from
nuclear
power].
I
use
the
word
'insane'
purposefully
since
one
of
its
definitions
is
loss
of
contact
with
reality.
The
public's
understanding
of
radiation
dangers
has
virually
lost
all
contact
with
the
actual
dangers
as
understood
by
scientist"
(35,
p.
31).
Risk
perception
research
paints
a
different
picture,
demonstrating
that
people's
deep
anxieties
are
linked
to
the
reality
of
extensive
unfavorable
media
coverage
and
to
a
strong
association
between
nuclear
power
and
the
proliferation
and
use
of
nuclear
weapons.
Attempts
to
"educate"
or
reassure
the
public
and
bring
their
perceptions
in
line
with
those
of
industry
experts
appear
unlikely
to
succeed
because
the
low
probability
of
serious
reactor
accidents
makes
empirical
demonstrations
of
safety
difficult
to
achieve.
Be-
cause
nuclear
risks
are
perceived
as
unknown
and
potentially
catastrophic,
even
small
accidents
will
be
highly
publicized
and
may
produce
large
ripple
effects
(Fig.
4).
Psychometric
research
may
be
able
to
forecast
the
response
to
technologies
that
have
yet
to
arouse
strong
and
persistent
public
opposition.
For
example,
DNA
technologies
seem
to
evoke
several
of
the
perceptions
that
make
nuclear
power
so
hard
to
manage.
In
the
aftermath
of
an
accident,
this
technology
could
face
some
of
the
same
problems
and
opposition
now
confronting
the
nuclear
indus-
try.
Placing
Risks
in
Perspective
A
consequence
of
the
public's
concems
and
its
opposition
to
risky
technologies
has
been
an
increase
in
attempts
to
inform
and
educate
people
about
risk.
Risk
perception
research
has
a
number
of
implications
for
such
educational
efforts
(36).
One
frequently
advocated
approach
to
broadening
people's
per-
spectives
is
to
present
quantitative
risk
estimates
for
a
variety
of
hazards,
expressed
in
some
unidimensional
index
of
death
or
disability,
such
as
risk
per
hour
of
exposure,
annual
probability
of
death,
or
reduction
in
life
expectancy.
Even
though
such
compari-
sons
have
no
logically
necessary
implications
for
acceptability
of
risk
(15),
one
might
still
hope
that
they
would
help
improve
people's
intuitions
about
the
magnitude
of
risks.
Risk
perception
research
suggests,
however,
that
these
sorts
of
comparisons
may
not
be
very
satisfactory
even
for
this
purpose.
People's
perceptions
and
attitudes
are
determined
not
only
by
the
sort
of
unidimensional
statistics
used
in
such
tables
but
also
by
the
variety
of
quantitative
and
qualitative
characteristics
reflected
in
Fig.
1.
To
many
people,
statements
such
as,
"the
annual
risk
from
living
near
a
nudear
power
plant
is
equivalent
to
the
risk
of
riding
an
extra
3
miles
in
an
automobile,"
give
inadequate
consideration
to
the
important
differences
in
the
nature
of
the
risks
from
these
two
technologies.
In
short,
"riskiness"
means
more
to
people
than
"expected
number
of
fatalities."
Attempts
to
characterize,
compare,
and
regu-
late
risks
must
be
sensitive
to
this
broader
conception
of
risk.
Fischhoff,
Watson,
and
Hope
(37)
have
made
a
start
in
this
direction
by
demonstrating
how
one
might
construct
a
more
comprehensive
measure
of
risk.
They
show
that
variations
in
the
scope
of
one's
17
APRIL
I987
definition
of
risk
can
greatly
change
the
assessment
of
risk
from
various
energy
technologies.
Whereas
psychometric
research
implies
that
risk
debates
are
not
merely
about
risk
statistics,
some
sociological
and
anthropological
research
implies
that
some
of
these
debates
may
not
even
be
about
risk
(5,
6).
Risk
concerns
may
provide
a
rationale
for
actions
taken
on
other
grounds
or
they
may
be
a
surrogate
for
other
social
or
ideological
concerns.
When
this
is
the
case,
communication
about
risk
is
simply
irrelevant
to
the
discussion.
Hidden
agendas
need
to
be
brought
to
the
surface
for
discussion
(38).
Perhaps
the
most
important
message
from
this
research
is
that
there
is
wisdom
as
well
as
error
in
public
attitudes
and
perceptions.
Lay
people
sometimes
lack
certain
information
about
hazards.
However,
their
basic
conceptualization
of
risk
is
much
richer
than
that
of
the
experts
and
reflects
legitimate
concerns
that
are
typically
omitted
from
expert
risk
assessments.
As
a
result,
risk
communica-
tion
and
risk
management
efforts
are
destined
to
fail
unless
they
are
structured
as
a
two-way
process.
Each
side,
expert
and
public,
has
something
valid
to
contribute.
Each
side
must
respect
the
insights
and
intelligence
of
the
other.
REFERENCES
AND
NOTES
1.
For
a
comprehensive
bibliography
on
risk
assessment,
see
V.
Covello
and
M.
Abernathy,
in
Technolical
Risk
Assesment,
P.
F.
Ricci,
L.
A.
Sagan,
C.
G.
Whipple,
Eds.
(Nijhoff,
The
Hague,
1984),
pp.
283-363.
2.
"Risk
in
a
complex
society,"
report
of
a
public
opinion
poll
conducted
by
L.
Harris
for
the
Marsh
and
McClennan
Company,
New
York
(1980).
3.
A.
Wildavsky,
Am.
Sci.
67,
32
(1979).
4.
I.
Burton,
R
W.
Kates,
G.
F.
VWhite,
Thc
Environment
as
Hazard
(Oxford
Univ.
Press,
Oxford,
1978).
5.
J.
F.
Short,
Jr.,Am.
Sociol.
Rev.
49,
711
(1984).
6.
M.
Douglas
and
A.
Wildavsky,
Risk
and
Cukure
(Univ.
of
California
Press,
Bcrkeley,
1982).
7.
W.
Edwards,
Annu.
Rev.
Psyol.
12,
473
(1961).
8.
D.
Kahneman,
P.
Slovic,
A.
Tversky,
Eds.
Judgment
Under
Uncertainty:
Heurstics
and
Bias
(Cambridge
Univ.
Press,
New
York,
1982).
9.
M.
Henrion
and
B.
Fischhoff,
Am.
J.
Phys.,
in
press.
10.
R.
Nisbett
and
L.
Ross,
Human
Inference:
Strtgs
and
Shortcomxgs
of
Social
Judgment
(Prentice-Hall,
Englewood
Cliffs,
NJ,
1980).
11.
A.
Tversky
and
D.
Kahneman,
Scence
211,
453
(1981).
12.
B.
Fischhoffet
al.,
Policy
Sci.
8,
127
(1978).
13.
P.
Slovic,
B.
Fischhoff,
S.
Lichtenstein,
Acta.
Psychol.
56,
183
(1984).
14.
C.
Starr,
Scence
165,
1232
(1969).
15.
B.
Fischhoff,
S.
Lichtenstein,
P.
Slovic,
S.
L.
Derby,
R.
L.
Keeney,Aceptabk
Risk
(Cambridge
Univ.
Press,
New
York,
1981).
16.
G.
T.
Gardner
et
al.,
J.
Soc.
Psycbo.
116,
179
(1982).
17.
D.
R.
DeLuca,
J.
A.
J.
Stolwijk,
W.
Horowitz,
inRiskEvaluation
andManoaement,
V.
T.
Covello,
J.
Menkes,
J.
L.
Mumpower,
Eds.
(Plenum,
New
York,
1986),
pp.
25-67.
18.
E.
J.
Johnson
and
A.
Tversky,
J.
Exp.
Psych.
Gen.
113,
55
(1984).
19.
M.
K.
Lindell
and
T.
C.
Earle,
Risk
Anal.
3,
245
(1983).
20.
H.
J.
Otway
and
M.
Fishbein,
The
Determinants
of
Attitude
Formation:
An
Application
to
Nudear
Power
(RM-76-80
Technical
Report, International
Institute
for
Applied
Systems
Analysis,
Laxenburg,
Austria,
1
21.
0.
Renn
and
B.
Swaton,
Env.
Int.
10,
557
(1984).
22.
P.
Slovic,
B.
Fischhoff,
S.
Lichtenstein,
in
Societal
RiskAssessment:
How
Safc
is
Safe
Enough?,
R.
Schwing
and
W.
A.
Albers,
Jr.,
Eds.
(Plenum,
New
York,
1980),
pp.
181-216.
23.
C.
A.
J.
Vlek
and
P.
J.
Stallen,
Organ.
Behav.
Hum.
Perf
28,
235
(1981).
24.
D.
von
Winterfeldt,
R.
S.
John,
K.
Borcherding,
Risk
Anal.
1,
277
(1981).
25.
P.
Slovic,
B.
Fischhoff,
S.
Lichtenstein,
in
Perlous
Pr
:
Managing
the
Hazards
of
Technoljy,
R.
W.
Kates,
C.
Hohenemser,
J.
X.
Kasperson,
Eds.
(Westview,
Boulder,
CO,
1985),
pp.
91-125.
26.
P.
Slovic,
B.
Fischhoff,
S.
Lichtenstein,
Environment
21
(no.
3),
14
(1979).
27.
Estimated
at
$500
billion
[see
Ekar.
Power
Res.
Inst.
J.
5
(iso.
5),
24
(1980)].
28.
P.
Slovic,
S.
Lichtenstein,
B.
Fischhoff,
Manage.
Sci.
30,
464
(1984).
29.
The
Talk
of
the
Town,
New
Yorker
60
(no.
53),
29
(1985).
30.
N.
Kraus,
thesis,
University
of
Pittsburgh
(1985).
31.
P.
Slovic,
D.
MacGregor,
N.
Kraus,AccidentAnal.
Prey.,
in
press.
32.
Grinshaw
vs.
FordMotor
Co.,
Superior
Court,
No.
19776,
Orange
County,
CA,
6
February
1978.
33.
P.
Slovic,
B.
Fischhoff,
S.
Lichtenstein,
inAdvances
in
Environmental
Psychology,
A.
Baum
and
J.
E.
Singer,
Eds.
(Erlbaum,
Hillsdale,
NJ,
1981),
vol.
3,
pp.
157-169.
34.
R
L.
Dupont,
Bus.
Week,
7
September
1981,
pp.
8-9.
35.
B.
L.
Cohen,Before
It's
TooLate:A
Scientiss
Casefor
NudearEnergy
(Plenum,
New
York,
1983).
36.
P.
Slovic,
RiskAnal.
6,
403
(1986).
37.
B.
Fischhoff,
S.
Watson,
C.
Hope,
Poly
Sci.
17,
123
(1984).
38.
W.
Edwards
and
D.
von
Winterfeldt,
Risk
Anal.,
in
press.
39.
The
text
of
this
article
draws
heavily
upon
the
authores
joint
work
with
B.
Fischhoff
and
S.
Lichtenstein.
Support
for
the
writing
of
the
artide
was
provided
by
NSF
grant
SES-8517411
to
Decision
Research.
ARTICLES
285
on February 1, 2010 www.sciencemag.orgDownloaded from