Risk Analysis, Vol. 20, No.

1,
2000

Factors in Risk Perception

Lennart Sjberg
1

Risk perception is a phenomenon in search of an explanation. Several approaches are

discussed in this paper. Technical risk estimates are sometimes a potent factor in accounting
for
perceived risk, but in many important applications it is not. Heuristics and biases,
mainly
availability, account for only a minor portion of risk perception, and media contents have
not
been clearly implicated in risk perception. The psychometric model is probably the
leading
contender in the eld, but its explanatory value is only around 20% of the variance of
raw
data. Adding a factor of unnatural risk considerably improves the psychometric
model.
Cultural Theory, on the other hand, has not been able to explain more than 510% of
the
KEY
WORDS:
Risk perception;
cultural
theory;
psychometric
paradigm
variance
of perceived
risk, and
other
value
scales have
similarly failed. A model is
proposed
in which attitude, risk sensitivity, and specic fear are used as explanatory variables;
this
1. INTRODUCTION
model seems to explain well ever,
over 3040%
of the
variance
and
is thus that
more risk
promising
Starr soon
followed
and
showed
than
accepprevious
approaches.
Theof
model
offers
different
type
psychological
explanationofof
tance
wasarelated
not
only
to technical
estimates
Perceived risk has been
a focus
of interest
and
benets
but also
to
aofsubjective
dimension
risk
polrisk as
such
voluntariness.
Although Starrs choice of data
perception,
and it
has manywas
implications,
e.g.,
a different
approach
to the
icymakers and researchers
for some
decades.
bases
criticized
work
was very
seminal
in relationship
(5) his
beRisk
up a new area
of research. The question of
(6) opening
tween attitude and perceived risk, as compared with the usual cognitive analysis of
perception appears to hold a central position in perceived
attitude.

the opposition to technology.

and
For example, a risk had thus been posed, and social scientists
political
study
agenda
of many countries
and is crucial rose to
of
private
bills submitted
to the Swedish
for
Parliament
the challenge of explaining this curious
(1,2)
the understanding
showed
that
the
of involvement
of risk-related
in(3)the
bills had
phenomenon.
2.
TECHNICAL
ESTIMATES
OF proposed
RISK
during
the
last
30share
years,
from 11 to
Risk
perSeveral factors
have been
for the
environment
tripled
29%.
ception
is an especially important aspect in the
explanunation of perceived risk. A primary candidate is,
clear eld.The present paper addresses the
of
question:
course, real risk. Although risk is always
Why do people perceive risks as they do?
construed,
The beginning
of risk perception
can real risk is a useful concept for expertly, or
devised
risk comparisons
which for aresearch
while were
be- be
otherwise
traced to
lieved
to be
thepertinent
nuclear for
debate
risk of
communication.
the 60s. SowbyHowwell-founded, assessed risk. Most authors have
1
had
tended
Stockholm School of Economics, Center for Risk Research, Box
(4)
to downplay
this factor,
for various
reasons.
Still,
ple,
in a well-known
paper
it was shown
that
6501, 11383 Stockholm, Sweden.
average
some
1 studies suggest
0272-4332/00/0200-0001$16.00/1
for Risk
that real
(7)
risk is
a 2000
verySociety
important
Analysis
determinant of perceived risk in some contexts. For
exam-

Sjberg

estimated mortality rates for a number of common

It can be added that the work on heuristics
illrelied
nesses and accidents were strongly related to
greatly on probability calculus problems that
statistical
had
data.There was also, to be sure, some systematic been set up to have counterintuitive answers.
deviHowation from the trendsmall risks were
ever interesting such a method may be for
overestimated
revealing
Others
have replicated these
and The
large
risks
the can
how intuition differs from probability calculus, it has
ndings.
type
ofwere
risksunderestimatedbut
where realistic perception
general
probably
limited applicability to how people
(8,9)
be
expected
appear to be the risks with which
4.
RISK TARGET
trend some
was the
one just stated:
risks
were In perceive
people
have
experience,
direct or
indirect.
perceived,
inthe
applied
addition,
it must be noted that the risk ratings of the probabilities
People doand
notrisks
make
samesettings.
estimate when
by the average person, in a rather veridical
cited
they
(10)
manner.
work
concerned general risk, i.e., the risk to others,
rate the risk to themselves, to their family, or to
or
people
to people in general. The risk target is a factorin general. For example, in a study with a large
of
repre- The respondents were instructed to rate
ards.
great importance in risk perception, to be treated in
sentative
sample
of 0the
the
risks
on a scale
from
(no Swedish
risk at all)population,
to 6 (an
a HEURISTICS, BIASES, AND RISK
these
extremely
(21)
3.
subsequent
section.
three risk),
large
risk targets
with ve
were
categories
studied for
lying
a set
in between
of fteen
PERCEPTION
hazthese
Whether people perceive a risk the same as two extremes and dened as 1, 2, 3, 4, and 5.The
mean
technical risk estimates can be seen as a question of ratings (N 1,224, 72% response rate) are given
the
in
(11)
veridicality
of subjective
was
Fig.1.Drastic differences are seen,particularly
and
probability
judgmentprobability.It
biases
was often cited
conceived Of the three heuristicsrepresentativebetween
pertinent.
in thatavailability,
way
in theand
1970s,
andas
the work
on often
general risk and personal or family risk.
ness,
anchoringit
was most
(12)
(13)
heuristics
Note that the difference is not only a question
arof
gued
ing
risk
that
perception.
availabilityThere
wasisimportant
an obvious
for
level. The rank order between risks is not the
relationship
understand(14)
here to mass media and the idea that frequent
same
media
across targets.
exposure gives rise to a high level of perceived risk.
is clear
not unrealistic
everybodyoptimism.
can be right in
whatIt has
beenthat
called
This work is no longer regarded as of primary It
sayis also important, of course, to
(22) ask what
importance for risk perception, however. Fischhoff
ing that
can he or she is subjected to a smaller risk
(15)
and others
showed
earlyone
on that
the publics
risk cause
than this variation in perceived risk across risk
tive
probability
is only
of many
factors.
The
per- media role in risk perception is still very
people in general (even if some respondents of
tarmass
ception was(16,17)
much more multidimensional
course
gets.
Respondents have also been asked to rate
much
subjecare right in saying so).The fact that people most
the
under
The
debate.
heuristics tradition relied heavily on a preoften
sumption that belief distortion is a matter of
account
cogni- for in a purely cognitive framework, andclaim to be less subjected to risk than others can
it
tion.It has strongly.
correlate
been (18,19)
found,however,that beliefs and be
is hard
suggests
that beliefsThis
are phenomenon
value driven. The
biasto
of termed risk denial.Risk denial is a very important
values
befealiefs is largely a question of a tendency towards ture of risk perception.The phenomenon is related
blackto
and-white construals and wishful thinking.
Montgomery pointed to the pervasive tendency of
decision
makers to
options
in such a manner as to Fig. 1. Average risk ratings: personal, family, and general
justied
toconstrue
oneself and
others.
risk.
ob(20)
tain a dominant or quasi-dominant alternative;
the
choice of such an alternative is of course very
easily

Factors in Risk Perception

Fig. 3. Mean ratings of 27 hazards plotted for different

targets;
independent groups rated each target.

An unpublished masters thesis from the Center

for
Fig. 2. Risk denial (general minus personal risk) plotted
Risk Research reported a study of perceived risk
against
and to sell wine.
The correlations between
perceived control over risks. Each point corresponds to one stores
hazthe attitude
to legislation
enabling ordinary
perceived
personal
risk of alcohol-caused
injury and
ard; mean ratings are plotted.
grocery
the
(25)
degree of control they felt they had over each of corresponding general risk on the one hand and
attithe
tude toward liberalization of sales on the other
hazardsto
what
they couldbetween
protect
selves
against
it. extent
Mean differences
themgeneral
and
personal risk (risk denial) are plotted in Fig.were
2
0.59 for general risk and 0.15 for personal risk.
against mean (23)
control ratings.
(In
Hence, at this level of analysis, control is
both cases, a larger risk was associated with a
apparently an important variable in accounting for risk more
negative attitude.) Hence, policy attitude was
denial.
For individual data, the correlation is much weaker.probably driven by general risk rather than personal risk.
In many risk perception studies the target is not
The risk target is of paramount importance in
explicitly spelled out. The respondents are
risk
studies. This fairly simple fact seems not to
simply
be
askedgroups
to rate the
risk,
with no further
into
getting
different
instructions.
generally known
many studies
interesting
factor. and
Although
it is truestill
thatwork
specication.
They
with
studying
the
choice
of
target
in
a
situation
with
ambiguous
ThatItinstruction
compared
with several
(24)
others
was foundwas
that
the no-target
condition
nonspecied risk
or
(26)targets.This means that they
in gave
rated
27 hazards.The results are seen in Fig. 3.
probanonspecic
instruction
as to target could be
a studythe
about
of same
309 subjects
result aswho
rating
were
the
divided
generalatrisk.
bly
interestmiss
out
on
the
need
to understand perceived
random
Furpering,this
is
simply
not
possible
in a design where
thermore, rating the risk to any one person
(26)
sonal
people
risk,
and
it
introduces
some
uncertainty as
Marris
et
al.,
also
is needed.risk ratings. Additional
to
just
make
nonspecic
gave the same result, refuting the sometimeswhat target they actually do study.Some authors
heard
Personal risk and general risk have different inforargue
mation
about the choice
of target, not collected
5.
THE PSYCHOMETRIC
MODEL
explanation
conof the generalpersonal difference that
explicitly
in
against
the
introduction
of a target and
it
sequences.
by way
of illustration,
attitudes
toConsider,
alcohol sales
liberalization
in a study
seem
This
model
was
launched
in
a
1978 paper by
reects the fact that general risk ratings refer to
of
Sweden.
(27)
to
regard
the
choice
by
the
subject
as to target as
Fischhoff
et
al.
2
many
2
At present, alcoholic beverages can be sold only by state
The empirical work reported in that
an
was
later
followed
up
with
several
other
studies
monopoly
stores,persons
which arethan
rather
few and risk
keepratings.
quite restricted
more
personal
paper
that
business
hours.Taxation keeps prices very high.The system has been inwere more extensive, both in terms of scales and
place
the
since the 1950s, when it replaced a still more restrictive
number of respondents.
system.

Sjberg

The model is based on a number of

with nature (tampering with nature, immoral
and
toryexplanascales (9 to begin with, later 18) such as
unnatural risk), which will be further
New
Old, VoluntaryInvoluntary, etc. The subjects are
described
asked to rate a number of hazards on each of subsequently.
the
3. The nding that a very large share of the
scales. Mean ratings are computed for each
varimean
ratings, not raw data. Mean data are less
hazard
ancetooferror
perceived
risk
could
be explained by
subject
than raw
data
andsmooth,hence
on each scale, and the resultant Scales
these
models
(31)
Hazards
factors
of
various
is due
kindsto
are
the
much
fact more
that likely
the authors
to t.One
matrix is factor analyzed. Usually, three
analyzed
could,
factors
of course, claim that mean risk ratings are the focus
sufce to explain a large share of the variance, of
perinterest, not raw data, which reect more directly
haps around 80%. The factors that have
the
been
risks actually perceived by the subjects. However,
foundoninrisk
many
studies (28)
are NewOld, Dread, and
for
work
communication.
The famous map
Number in
of the
Exposed.
hazards
two major dimensions of Dread practical and theoretical purposes it would seem to
of
usually account be
and The scales, or the factors,
NewOld
obvious that the individual ratings should be of
in a for
position
where
e.g.,
it ishow
highly
nuclear
loaded
power
on the
is Dread
(29) shows,
some 6070%
the seen
variance
of perceived
risk or 30%
primary
dimension
located
andofalso
as New
or Unknown.
by the traditional psychometric (32,33)
and
risk
interest.The
analysis
ofother
behavior
should
not
be
model,
This
this
means that
many
factors
are just
as of
acceptance,
when seemed
entered to
asbeaverage
ratings imporavernding
apparently
a self-evident
for a
age behavior
tant
as,devised
or more
but
of
the ratings
the people
explafactors
by important
Fischhoff
etthan,
al. that
number
of
hazards
in multiple
regression
analyses.
actually
psychometric
nation
Thisof
is the
curious
strong
indeed
public
because
opposition
no data
to
For one such dimension,(27)
consider results
This model has been used as a basis for
give.And
actually
show- for a practitioner, the inated levels of exnuclear
(34)
linked
extensive
opposition
to nuclear
power
to the
planatory
ing
that Risk,
a power
fourth
given
bydenoted
the results
as
Unnatural
reported
power
and is still often
quoted
as such.
Immoral
has factor,
been discovered
in recent
work.
psychometfor work is a continuation of earlier studies in
and
This
(35,36)
ric dimensions.Apparently it was sufcient that it which
the analyses of means simply
are very
just
misleading.
accounted
for neither by dread nor newness. the
moral factor was emphasized
and shown to
(37)
seemed
to be the
explanation.
As a
The raw data can only be explained to about
Indeed,
nuclear
power
is not perceived
as new. be
counterexample,
20 the reason
Per(30)
strongly
and
relatedmight
to perceived
be that risk.
hazards
Morality
are is indeed
the perceived
haps
the 15 years
risk of
since
nuclear
the rst
wasteU.S.
in Swedish
studies data
conceived
as much of the work on risk perception,
missing from
is
account
such, rather than as the outcome of human actions.
for that. But opposition to nuclear power is
In a study of perceived nuclear waste, risk
ever
data
present
today, just
it wasis at
the
the 541).
were obtained
from
a representative
sample of
Furthermore,
theasmodel
not
at end
all soofpowerful
A factor
analysis
of 21 psychometric
1970s.
the
scales
as
its proponents have claimed. Several reasons applied
to nuclear waste risk gave a four-factor
One wonders just how important the factor of
Swedish
population, response rate 53.2% (N
solu(34)
for
new-1.

with 66% explained variance. Oblimin rotation

this
statements
Factor analysis
can beofgiven:
a matrix with only 9 or tion
nesseven
is, or was.
resulted in four factors, three of which were
18 scales is bound to give few factors. The fact tradithat
tional, namely New Risk, Dreaded Risk, and Number
these few factors account for some 80% of the of Exposed. A fourth factor also emerged, loaded in
varisuch items as Unnatural Risk, Immoral Risk, and
ance of the scales is not surprising and does not atHuman Arrogance. A multiple regression of perall
ceived nuclear waste risk (a pooled index, no
imply
can
alsobased
account
the
target
2. that
The they
scales
were
on aforcompilation
Only the factor of Unnatural and Immoral
perceived
specied)
these
four factors
gavethe
the
of
Risk
had aagainst
signicant
value.
Note that
(5)
level which
factors
of suggested
riskinto
same
in the
extent.
earlier
Therisk
latter
factor
of psychoresults
70s,
turn
was
stimulated
byliterature
Starrs
could
the
shown model
in Tableperformed,
I.
work.
Although
18 scales may seem like a lot, they
metric
in its three-factor
be unrelated to the scales, and the scales could still
missed
version,
be least one important aspect, namely,
at
at the usual 20% level and that its performance
well described by 3 factors.
interference
was
strongly improved by the introduction of the
new

Factors in Risk Perception

5
The fourth factor throws quite a different light
on perceived nuclear risk than the three
Table I. Multiple Regression Analysis of Perceived Nuclear
traditional
than raw data, i.e., correlations reported were
Waste
factors.
based It is no longer a question of new or
Risk as a Function of Psychometric Explanatory Factors
dreaded
on means and therefore very high, thereby giving
Value
risk,
an but a more elaborated perspective having to
Explanatory variable
(four factor model)
do
impression that virtually all of risk perception was
Many exposed to risk
0.090
with
ac- notions about tampering with nature and
New risk
0.044
moral
counted for.
Unnatural and immoral risk
0.452
questions. Other current concerns such as genetic
Dreaded risk
0.006
enR 2adj three traditional factors
0.200
R 2adj four factors
0.280
gineering and the BSE (mad cow disease)
come
easily to mind in this connection and the fourth
fourth factor, which then carried the whole of its factor
extherefore seems worthy of serious consideration
planatory
in
Could power.
these ndings be replicated in a different
setting and with different hazards? Another set further risk perception work.
It is sometimes argued that scales of the
of
psychodata analyzed here are from a current
metric type are trivially related to risk perception
European
Union project called RISKPERCOM in which 798 and
respondents (response rate 61.0%), representativethat semantic overlap is the true reason for success
in
of
the Swedish population, answered an extensive accounting for risk perception. Items that have
the
risk
perception questionnaire. Results are given for same or very similar denotation can of course be
expersonal and general risk of domestic nuclear power pected to correlate quite strongly. In the
present
and
context, however, this argument is
factor scores of four factors (explaining 61% for the
variance)
of psychometric
scales
II.
The reason
for the rather
lowmeasuring
values of the
R adjexaggerated.
is
Level of perceived risk is not strictly
perprobably
that a rather specic psychometric
target
semantically
ception of a Chernobyl-type accident. Twenty-two
2
was
studied,
namely
A Chernobyl-type
overlapping with, say, newness or immorality of
scales
were once
more
compressed toaccident.
four
The conclusion drawn about the psychometric
a
factors:
model
is that,
in itsRisk,
traditional
three-factor
form,
New Risk,
Dreaded
Many Exposed,
and
Un-hazard. The fact that properties of a construct or
it
obnatural and Immoral Risk (including risk as a warnexplains
onlyto
a come).
rather modest
share
of given
the variance
ject are statistically related to a global
ing of worse
The results
are
in
6. CULTURAL THEORY OF RISK
of
assessment
Table
PERCEPTION
perceived risk. The widespread credibility that such as attitude or level of risk may be seen as
the
likely
The Cultural Theory of risk perception
Table II.apparently
Regression Analyses
Personal
of but is not an artifact. Yet, many researchers
model
enjoys of
seems
toand
be General
based Risk
largely
(38)
wasout in a book by Douglas and Wildavsky
spelled
Domestic
Nuclear
Power
as
a
Function
of
Four
Psychometric
on
voice
2
and made operational for quantitative study
Factors, and R
adj for the Three Traditional Factors
the fact
that analysis
was done on averages their interest in more distal explanations of
by
Dake.
rather
risk,
Value
types(39,40)
of people:
The theory
egalitarian,
species
individualistic,
that there are
i.e., in constructs that are contentwise less
hierarchic,
four
Personal risk
General risk
obviously
and fatalistic. These types of people will choose
related. A second major attempt at explaining
Many exposed to risk
0.016
0.054
to
New risk
0.000
0.013
risk Egalitarians: technology and the
be concerned with different types of hazards:
Unnatural and immoral risk
0.436
0.419
perception
is provided by Cultural Theory, and
environment
Dreaded risk
0.014
0.000
that Individualists : war and other threats to
R 2adj three traditional factors
0.086
0.094
theory
the tries to deliver such nonproximal
R 2adj four factors
0.199
0.198
explanamarkets
toryconcepts
of risklaw
perception.
Hierarchists:
and orderA discussion of
that Fatalists: with none of the above
theory and some related approaches of a distal
nature follows.

Sjberg

These four types of people were originally

vised form of the scales, have shown that the
sug-by a group-grid analysis. The potent
scales
gested
(46:26)
account
for about 5% of the variance, on
social
age.
Correlations with single items have turned
contextin an abstract and formal senseof a out
averto give a very low amount of explained
pervariance
(37,47)
son is assumed to be governing his or her beliefs.Aof
when
to other
explanatory constructs in
riskadded
perception.
full
models
Yet the pattern of relationships with social con(26,45)
empirical and quantitative investigation of the
cerns observed by Dake is apparently robust.
The
theory
question is what to conclude from that nding of
demands a study also of these social phenomena
a
no (41)
weak but systematic pattern of relationships.
gist.
Boholms conclusions
asbeen
to the
theoretical Several
suchofinvestigation
has
so far
published.
tus
the theory are
quite
negative.The
present
staThe
theory is a rich conceptual structure and authors have suggested that much stronger
paper
will concentrate
has
on what can be said about the
relationseveral variations; see Boholms critical analysisships are likely if the scales are improved.
empirof support claimed for the theory in
ical
However,
found
in Sjberg.
They were
(42)
the theoryisfrom
quantitative
the standpoint
of an
correcting
for scales
measurement
error yields
only
moderate.)
discussion
available
The Dake
are examples
of scales
of trivial
anthropolostudies
the relationship
between
the theory.
four types
increase
correlations. (Reliabilities of the
elsewhere.
politi-of the(45)
First,ofhowever,
a few words
about
The
of
U.S.values and attitudes,and they function much in
success
of Cultural Theory is largely an example cal
people noted previously and risk perception.A fuller
scales are not reported in Dakes work but can
the
of
be
same
manner as traditional political attitude
the persuasive power of speculation. The
scales
theory
(43)
has now
tial.
existed for some 15 years and it has when related to risk perception, i.e., quite weakly.
Yet, the only empirical evidence offered
Asking if value scales in general are
been, Dakes work was available in anecdotal
before
acin the words of a supporter, immensely
tion.similarly
Lifestyles are extremely popular in
counts,
weakthere
in accounting
for risk
perception
is
marketing
inuen- a few anthropological observations, and
are many claims
about
their strong
and
natural.
explan(48)
some observations using more or less
Sjberg
atory
value
rst approached
in various commercial
this questioncontexts.
by
(44)
informal
investigatHowThey used
qualitative
methods.
ing lifestyle
were
scales
found
and
torelating
carry
little
them
or no
to risk
scales
measuring
the four concepts illustrated by ever,they
A second
approach,
investigating
some
percepexplanatory
the The initial empirical support for the theory was
curdescribed
types
notedbypreviously
Wildavsky
andand
variously
Dake.termed
power for
rently
popular
risk perception.
general value scales in the
cultural
lated
with a number of social concerns, which environbiases orbeworld
Theofscale
scores were
mental eld, is summarized here.To achieve this
can
perhaps
seen views.
as a kind
risk ratings,
and
correpurreports
3
were given for a selection of the concerns that pose, a number of value scales were included in
were
a
signicantly correlated with the scales.The patternquestionnaire aiming at the measurement of
of
attitudes
relationships provided by these correlations to Swedish membership in the European Union.The
was
study was conducted during the month before
roughly in accordance with predictions from the the study, and representing a broad selection
the-However, the same scales and concerns
Swedish toreferendum
(November and
13, economy.
1994). A
of
threats
health, environment,
ory, were
but only bivariate correlations were given, They
mail
4
(45)
so
investigated
in a later study of U.S. subjects were
surveyjudged
designonwas
a category
used. Sixty
scale
percent
with of
7 steps,
a
Sjberg,
in which
explicit
risk ratings
were also 06,
judging Ithow
by
much
the variance
perceived
random
tained.
was
foundof that
concernsofwere
obconsample 0of denoted
where
the adulta Swedish
nonexistent
population
risk and 6 an
explained,
cerns
on
average,
would to
have
about
been
10%
explained
of the variance,
by multiple
risk
responded,
exregresratings
thus yielding
tremely
large arisk.
data set with 1,224 respondents.
3
Although this research has been suggested as strong evidence4 Personal and general risk perception correlate highly and have
sion
to
only
of
the
half
scale
as
much.
scores
European
is
impossible.
studies,
using
a
Twenty
The
value
two
scales
risks, all
were
personal,
Swedish
were
translations
included in
sifor
remilarofrelationships to other variables, so it is unlikely that
a Cultural Theory apporach to understanding opposition to
general
techthe
following:
nology, the social concerns actrully studied were mostly relatedrisk perception data would have been more strongly related
to
to
value dimensions.
economic and social problems, not problems of technology.

Factors in Risk Perception

Table III. Mean and Range of Explained Variance of 22 Personal Risks,

on the Basis of Background Variables and Value Dimensions
Explanatory variable
Sex
Education
Size of residential community
Income
Political party preference
Kahles list of values
Dunlaps New Environmental Paradigm
Ingleharts post-materialism
Schwartzs 14 general items
Schwartzs responsibility items
Schwartzs awareness of consequences
items
Kempton items
Left-right
political items

Minimum

Maximum

Average

.000
.042
.011
.000
.030
.007
.000
.010
.003
.000
.039
.008
.004
.163
.051
.000
.070
.024
.010 often display
.104
studies
a bias.046
of some kind,e.g.,in
.001
.095
.038
terms
of
.017
.129
.067
educational
level
of the respondents,
but results
.001
.093
.026
such
as
.000
.260
.058
.088(and they.035
the.000
present ones
are rather typical)
.005
.113
.050
suggest

that such bias is not serious in studies of risk

perception.
Several other objections to the critique of Cul(49,50)
1. Kahles
List of Values (8 items)
tural
Theory need to be discussed. It should rst
(51)
2. Dunlaps New Environmental Paradigm
be
(15 items)
noted that the results reported here are not all
3. Ingleharts postmaterialism items (12
that
(53)
4. Three subsets
(52) items)of items from
different from the ones reported by Dake.The differSchwartzs
value scale. These were 14 items selected
ences are mostly a question of the conclusions
on
drawn,
the basis of a previous study to cover the
not the results per se.
facDefenders of Cultural Theory have suggested a
tors of the scale in a representative manner,
number of objections to the present
9
conclusions,
associated with environmental
namely that the results are due to restriction of
(54)
responsibility
range
and 9 with consequences of
or differentiation (too few steps) of the
environmental
variables
risks, in all 32 items.
used, lack of reliability, the choice of risk
Multiple correlations (squared, adjusted)
5. Nine items from a study of American
perception
were
values
2
2
computed for each set of value
items
and eachrather than demand for risk mitigation, linguistic
by Kempton
adj or eta
per- for the risks and the following
puted
background dif6. Eleven items measuring current
left
were comculties or cultural differences;or it is argued that
ceived personal risk.In addition,R
data:
right
Sex
other
conict or traditional political ideology
Education
ways of making the theory operational could result
Income
in
Size of residential
research supporting the theory. None of these ad
InTable III,the mean amounts of variance
community
hoc
explained,
Political party preference
computed across the 22 risks, as well as minimumobjections holds up under closer
States than in Europe. The notion that the
and
scrutiny,however.The
theory
2
maximum values ofadjR
purely statistical objections (restriction of
or eta
5
These results show clearly
that general value range,lack of
2
may be supported in other, as yet untried, ways is
direliability, etc.) have been checked and found to be
are reported.
5
logThe left in Europe is more strongly associated with the
mensions are not very promising for
of
estab- correct but an uninteresting type of
ically
understanding
minor
importance.
differences
in political
lished power
and takesThe
responsibility
for technologies
such as
objection
risk perception. Even in the best case, only a
culture
nuclear power.
Hence, egalitarian
values cannot
be expected
modest
between
Europe
and the United
States
may be to
be
average of 67% of the variance was explained. the
as strongly correlated with antitechnology values as in the
It is also important to note that background reason
why the Dake scales are somewhat
United
factors
more
States.
were very weakly related to risk perception.
strongly related to risk perception in the
Survey
United

Sjberg

which can always be brought up when a theory

Third, consider the concept of specic fear. Any
fails.a more detailed discussion, see Sjberg.
For
hazard elicits thoughts about specic fearSumming up so far, neither Cultural
arousing
(55) Theory,
valelements.For example,the perceived risk of ying
ues in general nor the psychometric model
elicexplain
its notions about falling from a great height, or
much of risk perception.The latter model
burnaccounts,in
ing,or being killed by a violent explosion.Nuclear
its original form, for only some 20% of the
fear
Nuclear
variance
is associated with the specic fear of radiation.
Radiation, nonnuclear
of risk
perception.Adding
new dimension
Unnat- Consider the case of nuclear power and
Second,
consider the the
concept
of risk
Nonradiation, nonnuclear
ural sensitivity.
and Immoral Risk increases that gure
three
If
a set of risk ratings are correlated it is almost The
substancases
case
of hazards:
of radiation and nonnuclear is interesting.
altially but still leaves most of the variance
It
ways
unex- found that they correlate positively and
refers to a type of hazard that has an important,
rather
plained.
Theory
is an even
spe7.
RISK Cultural
SENSITIVITY,
ATTITUDE,
ANDless
strongly.
In turn,
successful
cic factor in common with nuclear hazards,
SPECIFIC
FEARthis suggests that a common
underattempt to explain risk perception than the
namely
lying
factorconcepts
is measured
by all
risktoratings,
psychoradiation. Therefore, an index of nonnuclear and
Three
will be
used
modelno
risk permatter
metric model.
Apparently
new and
ception.
Consider
rst the aconcept
ofquite
attitude, or radiation-relevant hazards was formed
what
type of hazard is being investigated. Two (excluding
different
af(56)
psyapproach
is needed.
fect.
tude
Conventional
is a function
of
attitude
beliefstheory
and values.
assumes that X-ray diagnostics from this index in the analyses
However,
chological
explanations
could
for this of
atti-argument
an
can be made
for account
the opposite
(57)
ndX-ray perceived risk reported subsequently).
direction
(57)
ing.
People could
differ
in to
risk
Results from this model for X-ray diagnostics
of inuence,
i.e.,truly
for attitude
besensitivity,
driving
i.e.,
beliefs.
tion.
In the present context, risk perception would and for domestic nuclear power are presented
(34)
some
persons
are verymodeling
upset and worried
about
here.
be
Structural
equations
supports
this ner.
are are
from
the same
study as
the one
virtucaused
by attitude, e.g., to nuclear power, not vestigating
TheseThe
twodata
cases
quite
interesting
because
nothe psychometric
model and
inally
the all hazards, whereas others are quite
they
natural
indifferent
other way around.
both(see
risk
involve
previous
ionizing
discussion),
radiation,but
withthey
an Nare
of still
and tranquil. Or people could have different
tech798.Tables
scalenologies
IV
and V give
perceived
the results
in quite
fordifferent
personalways.
and In
use habits. Some tend to use the high end of the general
Swescale,
den, respectively.
risk,
X-ray diagnostics is viewed in a quite
for some reason (perhaps linguistic habits),
positive
It is seen that the three explanatory variables
others
manner
ac- and nuclear power in a more neutral
tend to use the low end of the scale, no matter count
man- for a sizeable share of the variance of
what
perceived
hazard they rate. The latter hypothesis is,
risk.All values
were
highly
explained
variance
have
beensignicant.A
presented further
however,
analelsewhere.
weakened by the fact that correlations between ysis of the specic risk components is called
for.
(30)
Table IV. Regression Analyses of Personal Risk of Domestic
risk
The
Nuclear Power and X-ray Diagnostics, Personal Risk
ratings and different ratings tend to be lowa
high levels of explained variance reached are
nding
encourag value
not compatible with the notion that risk
ing but not high enough to suggest that an
Nuclear
X-ray
intercorrelaexhaustive
power
diagnostics
Explanatory variable
tions are due to variation between individuals analysis has been made. Models reaching above
Attitude
0.403
0.130
in
60%
0.201
scale-use habits. For the present model analysis, aSpecic radiation risk (nonnuclear) 0.216
General risk sensitivity (nonnuclear,
risk
nonradiation)
0.270
0.354
sensitivity index was used. It was based on a
0.445
0.313
R 2adj
rather
large number of ratings of risk of a non-specic
kind,
i.e., in a study of nuclear risk it would contain risk
ratings of nonnuclear and nonradiation hazards.

Factors in Risk Perception

perceived according to this tradition, which

coined
the very term risk perception. Here, risk
value
perception
Nuclear
X-ray
is a function of properties of the hazards. Even
power
diagnostics
Explanatory variable
if
many more properties than real risk are considered
Attitude
0.304
0.162
0.225
Specic radiation risk (nonnuclear) 0.297
in
General risk sensitivity (nonnuclear,
the model, a stimulusresponse kind of thinking
nonradiation)
0.259
0.336
is
0.459
0.335
R 2adj
still behind it.CulturalTheory,on the other hand,is
very
different. Here, risk perception is a reection of
The fact that attitude plays such a
the
role prominent
in the models is interesting. It suggests that social context an individual nds him- or herself
risk
in.
perception is to a large extent a question of
The reason why this approach fails is probably
ideology
that
in a very specic sense, not in the general sense the social context is construed in a very abstract,
that is
farposited by Cultural Theory. People who, for some
fetched manner, and that social context per se by
reason, are strongly in favor of nuclear power tendno
to
means is the sole determinant of risk perception, if
see it as
risk free,
viceare
versa.This
is a specic it
like:
beliefs
and and
values
often strongly
(58)
case(19)and
correlated
psychologically
Experts
has A
any
inuence
at of
all,interest
which remains
tothe
be seen.
further
theme
is that of
of the general
interdependent.
thatsee
people
tendas
tovery
see
in
nuclear principle
waste eld
its risks
If attitude is a crucial factor in risk perception,
consemostly
much
and the of
quences
present
risk perception.
results suggest
It is simplistic
this, then
just to
good properties
smaller
than theof
public
thosedoes;
concepts
they or
areobjects
also more
that aspercepthey
posdifference
between the two groups.
tion isthat
sume
largely
a high
anlevel
expression
of perceived
of specic
risk
carries
valueswith
portance,
but
space
precludes
a
detailed
like and
itive
to nuclear
mostly bad
power,
properties
and that
in explains
those that
part
they
of sion
(not here.
it
(30)
discusdisthe
general;(60,61)
demands
see
for Section
risk mitigation.
VI). RiskOther
communication
factors are of im8. CONCLUSION
would,
ACKNOWLEDGMENTS
with such a stance, require a very different
Risk perception is hard to understand. Several
approach
a study
the CECThe
project
factors inuence it, but some of the models
fromThis
thatisimplied
bywithin
other models.
wholeRISKliteraPERCOM
(Contract
Fl4PCT950016),
supported
suggested
ture on attitude change becomes relevant, andalso
this
by
for risk perception
explain
more
is the Swedish Council for Planning and Coordinasuasive
results, buthave
theyfailed
have to
been
based
on than
tion
ofextensive
Researchand
(FRN),
the Swedisheld.Also
Council for
a
averages
a very
well-established
Humanistic
and
Social
Science
Research
(HSFR),
rather
and
therefore
small fraction
quite misleading
of it. Some
as investigators
to the
other
the
havestrategy. Others have presented seemingly components enter risk sensitivity and specic
explanatory
tive
Nuclear Power Inspectorate (SKI), and
perapparently
power
of the
been
models.
satised
Practices
with statistical
such as these Swedish
fear.
the
Swedish
Radiation
signicance(59)
have
These are aspects
thatProtection
relate in aInstitute
natural (SSI).
way to
as a criterion
clouded
the view
of validity,
of risk but
perception
that is aand contributed
perREFERENCES
counterproducto
a premature closure. In this paper an attempt sonality and clinical psychology.
has
1. L. Sjberg, Risk Perception by the Public and by Experts: A
been made to show that much remains to be done Dilemma in Risk Management, Human Ecology Review 6,
19 (1999).
and
2. L. Sjberg, Policy Implications of Risk Perception Research:
some ideas about how better models of risk
A Case of the Emperors New Clothes? paper presented at
percepRisk Analysis: Opening the Process, organized by Society for
Risk AnalysisEurope, Paris, October 1114, 1998.
tion can be formulated have been suggested.
3. L. Sjberg,A. af Whlberg, and P. Kvist,The Rise of Risk: Risk
The model suggested in SectionVII implies a dif- Related Bills Submitted to The Swedish Parliament in 1964
ferent psychological functioning in risk
65
and 199395, Journal of Risk Research 1, 191195 (1998).
perception
than in other models.The psychometric model is
cognitive in its conception and avorrisks are
indeed
Table V. Regression Analyses of General Risk of Domestic
Nuclear Power and X-ray Diagnostics, General Risk

10

Sjberg

4. F. D. Sowby, Radiation and Other Risks, Health Physics 11,

27. B. Fischhoff et al.,How Safe Is Safe Enough? A Psychometric
Study of Attitudes Towards Technological Risks and Benets,
879887 (1965).
5. C. Starr, Social Benet versus Technological Risk,
Policy Sciences 9, 127152 (1978).
6.
H. J. Otway and J. J. Cohen, Revealed Preferences: ComScience
28. P. M. Sandman, Responding to Community Outrage:
ments
On The Starr
Benet-Risk Relationships, Research
Strategies
165, 12321238
(1969).
Memorandum, 75-5, IIASA, Laxenburg,Austria (1975).
for Effective Risk Communication (American Industrial Hy7. S. Lichtenstein et al., Judged Frequency of Lethal
giene Association, Fairfax,Virginia, 1993).
Events,
29. P. Slovic,Perception of Risk, Science 236, 280285 (1987).
Journal of Experimental Psychology: Human Learning and30. L. Sjberg and B.-M. Drottz-Sjberg, Risk Perception of NuMemory 4, 551578 (1978).
clear Waste: Experts and the Public, RHIZIKON: Risk Research Report, 16, Center for Risk Research, Stockholm
8. D. von Winterfeldt, R. S. John, and K. Borcherding, Cognitive Components of Risk Ratings, Risk Analysis 1, 277287 School of Economics (1994).
31. G.T. Gardner and L. C. Gould,Public Perceptions of the Risk
(1981).
9. B. Vessby et al., Vra Dominerande Hlsorisker, Skrift 5, and Benets of Technology, Risk Analysis 9, 225242 (1989).
32. L. Sjberg,A Discussion of the Limitations of the PsychometRiskkollegiet (1992).
10. C. P. Thompson and D. Mingay, Estimating the Frequency of ric and Cultural Theory Approaches to Risk Perception, Radiation Protection Dosimetry 68, 219225 (1996).
Everyday Events, Applied Cognitive Psychology 5, 497
33. C. Marris et al., Exploring the Psychometric Paradigm:
510
Comparisons between Aggregate and Individual Analyses,
(1991).
11. A.Tversky and D. Kahneman,Judgment Under Uncertainty: Risk Analysis 17, 303312 (1997).
34. L. Sjberg, Risk Perceptions by Politicians and the Public,
Heuristics and Biases, Science 185, 11241131 (1974).
RHIZIKON: Risk Research Reports, 23, Center for Risk Re12. L. Sjberg, Strength of Belief and risk, Policy Sciences
search, Stockholm School of Economics (1996).
11,
35. L. Sjberg and E.Winroth,Risk, Moral Value of Actions, and
3957 (1979).
13. A. Tversky and D. Kahneman, Availability: A Heuristic for Mood, Scandinavian Journal of Psychology 27, 191
Judging Frequency and Probability, Cognitive Psychology 4, 208
(1986).
207232 (1973).
14. B. Combs and P. Slovic, Newspaper Coverage of Causes36. L. Sjberg and G. Torell, The Development of Risk Acceptance and Moral Valuation, Scandinavian Journal of Psycholof
ogy 34, 223236 (1993).
Death, Journalism Quarterly 56, 837843, 849 (1979).
15. B. Fischhoff, P. Slovic, and S. Lichtenstein, Lay Foibles 37. L. Sjberg and B.-M. Drottz-Sjberg, Moral Value, Risk and
Risk Tolerance, RHIZIKON: Risk Research Report, 11, Cenand
Expert Fables in Judgments about Risk,American Statistician ter for Risk Research, Stockholm School of Economics (1993).
38. M. Douglas and A.Wildavsky, Risk and Culture (University of
36, 240255 (1982).
California Press, Berkeley, 1982).
16. W. R. Freudenburg et al., Media Coverage of Hazard Events:
39. K. Dake, Technology on Trial: Orienting Dispositions toward
Analyzing the Assumptions, Risk Analysis 16, 3142 (1996). Environmental and Health Hazards, Ph. D. thesis, University
17. A. af Whlberg and L. Sjberg, Risk Perception and the Me- of California, Berkeley (1990).
dia.A Review of Research on Media Inuence on Public Risk 40. K. Dake, Orienting Dispositions in the Perception Of Risk,
Perception, Journal of Risk Research 3, 3150 (2000).
Journal of Cross-Cultural Psychology 22, 6182 (1991).
18. L. Sjberg, Beliefs and Values as Components of Attitudes,
41. . Boholm,The Cultural Theory of Risk:An Anthropological
in B. Wegener (ed.), Social Psychophysics, pp. 199218
Critique, Ethnos 61, 6484 (1996).
(Erl42.
Explaining
Riskand
Perception:
An Empirical
and
43. L.
S. Sjberg,
Rayner,Cultural
Theory
Risk Analysis,
in S. Krimsky,
baum, Hillsdale, New Jersey, 1982).
Quantitative
Evaluation
of Cultural
Theory,
Risk pp.
Decision
D. Golding
(eds.), Social
Theories
of Risk,
83
19. L. Sjberg and A. Biel, Mood and BeliefValue correlation, and
and
116 Policy 2, 113130 (1997).
Acta Psychologica 53, 253270 (1983).
44.
A.Wildavsky and K.CT,
Dake,Theories
of Risk Perception:Who
1992)
20. H. Montgomery, From Cognition to Action: the Search for (Praeger,Westport,
Fears What and Why? Daedalus 119(4), 4160 (1990).
Dominance in Decision Making, in H. Montgomery and O.
Svenson (eds.),Process and Structure in Human Decision 45. L. Sjberg,World Views, Political Attitudes and Risk Perception, RiskHealth, Safety
and Environment 9, 137152
Mak(1998).
ing, pp. 2350 (Wiley, Chichester, England, 1989).
46. J. Brenot and S. Bonnefous, Approche Socio-culturelle de la
21. L. Sjberg, Solstrlningens Risker: Attityder, Kunskaper och
Perception des Risques, Note SEGR/LSEES, 9517, FonRiskuppfattning, RHIZIKON: Rapport frn Centrum fr
Riskforskning, Handelshgskolan i Stockholm, 3. Centrum tenay-aux-Roses: IPSN, 1995.
47. L. Sjberg, Risk Perception by Experts And the Public,
fr Riskforskning, Handelshgskolan i Stockholm, 1994.
22. N. D.Weinstein,Unrealistic Optimism about Illness Suscepti- Rhizikon: Risk Research Report, 4, Center for Risk Research,
bility: Conclusions from a Community Wide Sample, Journal Stockholm School of Economics (1991).
48. L. Sjberg, Life-Styles and Risk Perception, RHIZIKON:
of Behavioral Medicine 10, 481500 (1987).
Risk Research Report, 14, Center for Risk Research, Stock23. L. Sjberg,AIDS: Riskuppfattning,Attityder och Kunskaper.
holm School of Economics (1993).
En Enktunderskning av ldersgrupperna 3045 r, RHIZ49. L. R. Kahle, S. E. Beatty, and P. Homer,Alternative MeasureIKON: Rapport frn Centrum fr Riskforskning, 1. Center for
ment Approaches to Consumer Values: The List of Values
Risk Research, Stockholm School of Economics, 1991.
(LOV) and Values and Life Style (VALS), Journal of Con24. L. Sjberg, Perceived Risk vs Demand for Risk Reduction,
sumer Research 13, 405409 (1986).
RHIZIKON: Risk Research Report, 18, Center for Risk Re- 50. L. R. Kahle, The Nine Nations of North America and the
search, Stockholm School of Economics (1994).
Value Basis of Geographic Segmentation, Journal of Market25. M.-L. Bernstrm and C. Lindholm,In Vino Veritas. En Studie ing 50 (2,April) (1986).
av Svenska Folkets Attityder till Systembolaget och en Friare
51. R. E. Dunlap et al.,Measuring Endorsement of an Ecological
Alkoholfrsljning, unpublished Masters thesis. Handelshg- Worldview:A Revised NEP Scale, paper presented at annual
skolan i Stockholm, 1995.
26. C. Marris, I. Langford, and T. ORiordan, Integrating Sociological ad Psychological Approaches to Public Perceptions of
Environmental Risks: Detailed Results from a Questionnaire
Study, CSERGE Working Paper, GEC 96-07. Norwich, University of East Anglia, Centre for Social and Economic Research into the Global Environment, 1996.

Factors in Risk Perception

11

meeting, organized by Rural Sociological Society, State 57. L. Sjberg, Psychological Reactions to a Nuclear Accident,
Paper presented at the Proceedings of a Conference on
Col52. lege,
R. Inglehart,
Culture Shift
in Advanced Industrial
The
Pennsylvania,August
(1992).
Radiological and Radiation Protection Problems in Nordic
Society
Regions, Troms 2122 November, 1991, Paper 12, J. Baarli
(Princeton University Press, Princeton, NJ, 1990).
53. S. H. Schwartz, Universals in the Content and Structure (ed.) (Nordic Society for Radiation Protection, Oslo, 1992).
58. L. Sjberg and H. Montgomery, Double Denial in Attitude
of
Values: Theoretical Advances and Empirical Tests in 20 Coun- Formation, Journal of Applied Social Psychology 29, 606621
54. tries,
W. Kempton,
J. S.
and J. A.
Hartley,
Environmental
59. (1999).
F. L. Schmidt, Statistical Signicance Testing and
Advances
in Boster,
Experimental
Social
Psychology
25, 1
Values
in American Vulture (MIT Press, Cambridge, 1995) . Cumulative
65
55. (1992).
L. Sjberg, Explaining Risk Perception: An Empirical and Knowledge in Psychology: Implications for Training of ReQuantitative Evaluation of Cultural Theory, Rhizikon: Risk
searchers, Psychological Methods 1, 115129 (1996).
Research Reports, 22, Center for Risk Research, Stockholm
60. L. Sjberg,Why Do People Demand Risk Reduction? paper
School of Economics (1995).
presented at the ESREL-98: Safety and Reliability, S.
56. M. Fishbein and I.Ajzen, Belief,Attitude, Intention, and
Lydersen,
61. G.
L.K.
Sjberg,
Perceived
vs Demand
Reduction,
BehavHansen,
and H. A. Risk
Sandtorv
(eds.), For
pp. Risk
751758
(Trondior:An Introduction to Theory and Research (Addison-Wesley, Journal
of Risk
Research
2 (in press).
heim:A.A.
Balkema,
1998).
Reading, Massachusetts, 1975).