230R94902
Communicating Effectively about Risk Magnitudes:
- - f
Bottom Line Conclusions
. j
and Recommendations for Practitioners
by Peter M. Sandman and Neil D. Weinstein1
1994
This is a brief summary of key conclusions and recommejndations from three
sequential pieces of research. For a longer summary of the Ph|ase Three research,
see the Executive Summary of Neil D. Weinstein, Peter M. Sandman, and William K.
, - .<''. .
Mailman, Communications to Reduce Risk Underestimation and;Overestimatjon
(January 1994). For a longer summary of the Phase Two research, see the Execu-
tive Summary of Neil D. Weinstein, Peter M. Sandman, and Paul Miller, Communicat-
ing Effectively about Risk Magnitudes, Phase Two (September ]I991). For a longer
i : _
summary of the Phase One research, see the Executive Summary of Neil D. Wein-
M. Sandman was director of the Environmental Communicatiorj Research program at
Rutgers University, and is now a risk communication consultant ai C4 Gray Cliff Road, Newton Center,
MA 02159 (617/630-0385), Neil D. Weinstein is'Professor of Human Ecology, Cook Office Building;
P.O. Box 231, Cook College, Rutgers University, New Brunswick, NJ 089031(908/932-9169). '
-------
2
stein, Peter M. Sandman, and Nancy E. Roberts, Communicatiiig Effectively about
Risk Magnitudes (September 1989).
All three" reports are available from the Risk Communication Project, Office of
Policy, Planning and Evaluation, U.S. Environmental Protection'Agency, or from the
Center for Environmental Communication, Rutgers University.
Probably the most important thing to say about the conclusions and recorri-
s , ' '
mendations that will follow is that they are preliminary. The effects we have found
i
are often small, taking a careful study with a big sample to find.; They are based on
people's reactions to hypothetical exposure data; we do not know if people respond
similarly to real exposure data. They are based on studies of several different risks
(radon, asbestos, etc.), but just a single health outcome, lung ckncer. And they are
based on just one or two studies; experienced social scientists know not to rely too
/heavily on a finding until it has turned up in several different .studies using several
i
different methodologies. In addition, participants in this research were much better
i
educated than the general population: Finally, the research design we employed
confronted people with personal .choices about an individually remediable pollutant in
I
their homes; a public risk controversy requiring large-scale mitigation by government
agencies might have generated very different responses.
!
Of course, practitioners can rarely afford to wait for definitive research results.
I *
Since you have a job to do, a risk to describe, you are better olf following the advice
below than ignoring it. But see it as tentative. !
i-
It is worth emphasizing that this research effort focused on ways of explaining
risk magnitudes more effectively that is, ways to help people understand the size of
' j
their risk. A more controversial class of risk communication strategies attempt to
influence risk responses by manipulating emotions, or behavior father than through
improved understanding (examples include dramatic fear appeals, social pressure,
-------
3
rewards for compliance, etc.). These non-cognitive approaches can be very effective
but many scientists object to them.
Seven Factors that Affect Risk Response
1. Data about the Actual Risk. With everything else held constant, subjects
' " - . ' '.''' ; '
viewed the risk as more serious when the data they were given told them the
actual probability of experiencing harmful effects was greiater. "(Henceforth in
this summary we will use the phrases "actual risk" and "actual probability of
experiencing harmful effects interchangeably.) j
In the Phase Two research, a ten-times-higher risk from geological
radon affected risk perceptions but not mitigation intentiojns; a 24-times-higher
risk from radon affected both. (The data provided included risk probability
information plus comparisons to smoking.) In the Phase Three research,
people facing a 40-in-1,000 radon risk expressed higher threat perceptions and
!
action intentions than people facing a 1-in-100,000 risk, .iln a more demanding
test, people facing a 40-in-1,000 risk still expressed lowejr threat perceptions
and action intentions than those facing a 400-in-1,000 risjk.
This is an encouraging bottom-line conclusion: Telling people the size of
the risk they face does help encourage an appropriate response. But large
differences in actual risk yielded modest differences in perceived risk and
action intentions. Moreover, the Phase One research shpwed that the effect of
data about the risk can easily be swamped by other factors, such as an action
standard or a risk ladder. And the Phase Three researclji showed that the risk
probability effect also tends to disappear when the actual risk is low and
"outrage" is high: Responses to nuclear power plant waste used in the con-
struction of home basements were no weaker the risk was 1-in-1,000,000 than
when it was 1-in-100,000. -';.
-------
4 _ ;- ; ' .
2. An Action Standard. The Phase One research found that formats that
included an action standard were superior to formats without an action stan-
dard in helping people respond in proportion to the actual'risk from radon or
asbestos. That.is, the relationship between actual risk aipd perceived risk, and
- . .' j ,.-..'
between actual risk and mitigation intentions, was stronger with an action
standard than without. The effect of providing an action 'standard, in fact, was
stronger than the effect of providing risk probability data.
People seemed to use the standard as an "anchor" to help them
interpret their own levels. The standard was especially pjowerful in helping
people distinguish levels above the standard from those below the standard
i '., ,
so powerful that it sometimes created an artificial discontinuity in risk percep-
tions at the standard. It also helped people distinguish kjvels just below the
standard from those far below the standard. On the other hand, an action
standard did not help people make the distinction between levels just above
the standard and those far above the standard. Three ranges seemed to be
psychologically meaningful when a standard was provided: "okay" (way below
the standard), "possible trouble" (just below) and "deep trouble" (anywhere
above the standard). !
. . f x
Practitioners should always provide a standard when one exists. Even
though the standard may tend to distort people's understanding of .the risk
(way below versus just below versus above), this trichotojmy is still closer to
, the actual risk than people's responses would be without] the aid of the stan-
i
dard. It may help to qualify the standard, where appropriate, with warnings
that risks just below the standard are nearly as risky as those just above, and
that risks far above the standard are much more risky th..
(An action standard without additional risk information is useful when an
apathetic response is anticipated and the goal is to provoke more risk aver-
* - t \
sion. See Number Four below.) j
-------
' .' . . - 5 \ ;' .
3. Advice. People said they felt less uncertainty and had a better under-
standing of their risk when advice was provided. More importantly, although
people m the Phase One study often said they would choose to mitigate at
levels below the standard, those receiving action advice showed this tendency
least. That is, adding advice to the standard made people less likely to "over-
react" vis-a-vis the standard, more likely to accept the rebornmendation not to
take action at low levels. Advice was not similarly useful at high levels; it did
not increase the probability that those above the recommended action level
- ' ' - '!-'
would plan to act (most already said they planned to act]).
Providing explicit advice is thus especially useful fbr panic prevention, to
deter overreaction at low risk levels. Its value for increasing remedial action at
i '
high levels (beyond what would be expected with a standard alone) was not
demonstrated. j
4. A risk ladder. People felt more at risk when presented simply with a
i ''''
suggested "action level" at which mitigation is recommended than when
' ' . ' ' !'
presented with such a standard located midway up a risk "ladder." In the
i .
Phase One research, thejadder included mortality data and risk comparisons;
in Phase Two it did not. In both studies, the context that the ladder provided
and the implication that levels higher than one's own are not rare appeared
i ' '
to reassure subjects and reduce their perception of risk, j In Phase Two, the
presence or absence of a risk ladder, even without any additional information,
had an effect on perceived risk equal to a several-fold difference in actual risk.
;j
Like a standard, in other words, a risk ladder actsjas a sort of "anchor."
Assuming the ladder goes higher than the individual's risk, level, the ladder
communicates that things could be worse. If the communicator's goal is
maximum risk aversion that is, if the hazard is serious [and the audience is
', ' ' i
inclined toward apathy a standard without additional injformatipn is ideal; its
. ' ' ' - }' . . ' ':
very ambiguity generates the desired risk-averse response. If panic is a
-------
. 6 ' '-:
problem,and the goal is to provide reassuring context, ori the other hand, a risk
ladder is worth adding. , '---:'
5. Location on the Risk Ladder. By modifying the risk ladder, the Phase
Two research was able to locate the same hypothetical Hazard exposure with
the same risk information either one-quarter of the way up the ladder or three-
j ,' .
quarters of the way up the ladder. The difference in location significantly
I
affected perceived risk in two experiments, arid mitigation intentions in one.
This locational effect was roughly equivalent in size to the effect of a 10x
i 'i
difference in actual risk. .;
In Phase Three, a risk chart was created that combined the effects of
!
the ladder (with location one-quarter or three-quarters of Ithe way up, as
appropriate) with a recommended action standard and ris'k comparisons.
Compared to the no-chart conditions, the chart increased; perceived threat
i
[ ' -
when the actual risk was high and decreased perceived threat when the actual
i ' ,
risk was low. The effect of the chart on risk perception was as great as or
; , ' /
greater than the effect of a 10x difference in actual risk, i The effect of the
chart on action intentions was smaller; it was still significant when the actual
risk was low, but not quite significant for the high-risk scenarios.
i
Risk information developed to guide laypeople is often arrayed on a risk
ladder, and the structure of the ladder may be determined more or less
' !,'"
arbitrarily. How low should the ladder begin? How high should it rise? Should
the scale be linear or logarithmic? The answers to thesej questions are not
obvious. What is clear from the data is that people's riskj perceptions can be,
' ' ' ' ! - '' .
substantially altered whether intentionally or inadvertently by constructing
the ladder so that their risk appears low or high on the page.
For helping people distinguish between high and low levels of a particu-
lar risk, X, the most effective ladder would be truncated ajt both ends, so that
high levels of X appeared at the top of the laddef and lovi/ levels of X at the
-------
7
bottom. For helping people see that risk X is actually lesjs serious than risk Y,
on the other hand, the ideal ladder would be extended upward, so that all
'' ' ' i - '
levels of X clustered near the bottom of the ladder, with V near the top. The
best ladder to help people see that X is actually more serious than Z would be
extended downward, clustering all the levels of X near th£ top, with Z near the
bottom. A "universal ladder" incorporating all three risks jwould extend both
upward and downward, and would cluster all the levels of X near the middle,
with Y near the top and Z near the bottom. These three iextended ladders
would all be improvements on the original truncated ladder in encouraging an
appropriate response to between-hazard risk differences j(X versus Y versus Z)
but they would all be worse than the original in encouraging people to
i
discriminate within-hazard risk differences (high versus Icjw levels of X). Thus,
it may be impossible to construct a risk ladder that makes optimal use of the
locational effect for all risk levels included on the ladder, i
6. Comparisons to Normal Background. The most surprising finding of the
Phase Three research was the powerful impact of comparisons to normal
background levels on threat perceptions and action intentions. Despite having
, ' I . '
no information at all about the likelihood of harmful consequences (that is,
about risk), people responded strongly to the information that their radiation
',j
exposure was either 20x higher than normal background] radiation or 200x
lower than normal background. In fact, the comparison to normal background
did a better job than risk information itself in helping people respond in proper-
-i- > .
tion to the actual risk. The effect of the comparison to normal was equal to or
greater than the effect of a 10-fold difference in actual risjk.
Comparisons to normal were especially powerful when risk was small
and the outrage substantial precisely the situation wheh risk information itself
was least powerful. Even more impressively, comparisons to normal seemed
- ' '! '
to affect the outrage itself. Subjects in the high-outrage "nuclear waste"
-------
' ' " . 8 - .'!"' ''' ':
conditions were understandably angry. Their anger was jreduced far more by
the knowledge that the situation posed a risk 200x less than normal back-
,i . '
ground than by the knowledge that the risk posed was ajmere 1-in-i 00,000 or
even a mere 1-in-1,000,000. j
In effect, people, may view the normal background
exposure as the
maximum safe exposure, regardless of the level of risk itj presents. This
i
exposure can then serve as an "anchor" for their risk judgments, with higher
levels seen as "unsafe" or "unacceptable" and lower levels as "safe" or
".''.''' . j
"acceptable." The potency of the comparison to normal background and its
symmetry (that is, its effectiveness in both low-outrage, high-risk situations and
high-outrage, low-risk situations) suggest that it may be a valuable piece of
information to include in a risk communication. j
'"'!' ;
Three qualifiers are needed on this advice, however. First, information
on normal, background levels is often not available. Second, the research
focused on situations where background was a small fraction .or a sizeable
multiple of the risk under discussion; it isn't clear how people might respond, to
,a non-natural risk about equal to natural background. Finally, it js important to
note that comparisons to normal background can be misleading. For some
' " !'-.'
hazards, normal background levels are sufficient to constitute a meaningful
health risk, and even a small increment would be unwise if it were preventable.
For other hazards, the risk due to normal background exposure is negligible,
and an exposure many times background would still be negligible.
7. Outrage. In the Phase Three research, outrage substantially affected
threat perceptions and action intentions. That is, subjects in the high-outrage,
low-risk situation reported much higher perceived threat and higher action
' if. ~ ''
intentions than subjects in the lowoutrage, low-risk situation/although the
actual risk was identical.
-------
9
When subjects received only risk numbers, the outrage effect was just
as lajge as the 4,ODO-fpId,(!) difference in risk between tlie;high-risk and low-
risk conditions. In other words, people had about the same threat perceptions
' i
and action intentions whether they faced a low-outrage home, radiation risk of
40-in-1,000 (from geological radon)'or a high-outrage hofne radiation risk of 1-.
in-100,000 (from nuclear waste).
When communication was improved by comparisons to normal back-
ground levels or by the risk chart, however, the outrage effect, though still
substantial, was smaller than the 4,000-fold difference in |risk. We are encour-
aged by this apparent ability of some kinds of risk information to reduce threat
perceptions and action intentions even in the presence of high outrage. Many
practitioners have suggested that when people are outraiged, explanations of
the risk data are unlikely to prove fruitful. In the Pha.se three research,
i". ' '"''
outrage certainly increased threat perceptions and actionj intentions but
outrage did not diminish the ability of comparisons to background and risk
charts to reduce threat perceptions and action intentions.!
!
1
Five Factors that May Not Significantly Affect Risk Response
Risk Comparisons. In the Phase One research, trje addition of risk
comparisons to cigarette smoking had two effects: it made people feel the
' - i - -
brochure was more helpful and they .understood their risk! better, and on some
measures it made them less risk-averse (for example, the comparisons raised
the highest level people would find acceptable). However, the comparisons
H -. " ' ! ' ' "
had no effect on people's ability to distinguish high risks from low risks: no
effect on the accuracy of illness probability estimates or on the relationship
-i
between actual risk and perceived risk or mitigation intentions. In other words,
in most respects the impact of risk probability data was n;ot improved by the
- \ ' ^ i
inclusion of comparisons to smoking risks. I;
-------
10
Risk comparisons may of course prove more helpful in ways not exam-
* , - - v. ' , . i
ined in this research different comparisons, different situations but the
research so far provides little guidance on how to deploy risk comparisons
' '"' f .
usefully. . ;
2. Graphical Presentation. A bar graph showing risk probabilities at
different exposure levels functioned in the Phase One research exactly like risk
comparisons. It improved people's ratings of the helpfulness of the brochure
and their certainty about their risk, and made them somewhat less risk-averse.
' - : . . . . "i ..
However, there were no significant differences between graphical and strictly
quantitative presentations of risk data in the extent to which people distin-
guished high levels from low levels or radon from asbestos. Graphical dis-
plays, in other words, did not strengthen the relationship between the actual
,'...'' ' i
risk and people's responses to that risk.
In Phase Three, a matrix of dots (to represent the denominator of a risk
probability fraction) and a matrix of dots and X's (to represent the denominator
i
and numerator, respectively) were tested for two hypothetical health risk
decision problems. Neither form of graphical presentation had any effect on
people's responses.
It is of course possible that different graphical devi'ces would show a
greater impact on risk response.
3. Magnitude of Test Numbers. The Phase Two research tested the
hypothesis that people respond to risk data in terms of the magnitude of the ,
test numbers themselves, quite apart from the risk represented by those
numbers. By expressing asbestos risk alternatively in fibprs per liter and in
fibers per cubic foot, a 30-fold difference in numerical magnitude was achieved
without any difference in risk (as presented in terms of probabilities plus
smoking comparisons). No significant effects of the magnitude manipulation
I
were found. " '
-------
11
This somewhat surprising finding is reassuring. Concentration levels for
radon in water, for example,,are typically much greater than for radon in air,
although the waterborne risk is usually lower. It is encouraging that homeown-
ers are apparently able to disregard the misleading test magnitude cue, at
i
least when mortality information and smoking comparisons are also provided.
. . .. j . ,
4. Simultaneous Presentation. An additional factor teisted in the Phase
Two research was the possibility that the simultaneous presentation of asbes-
i ' ' '
tos and radon risks on the same ladder might help subjects understand that
i
the asbestos risk was less serious than the radon risk. This hypothesis was
rejected. There were no significant differences between the joint and separate
presentations for either radon or asbestos.
Of course it is possible that a different use of simultaneous presenta-
tions might help owners take note of risk differences fo!r example, presenta-
tions that included the different action levels for the two hazards, or presenta-
tions that directed readers' attention to the differences more forcefully or
interactively. !
,5. Information Overload. One of the formats tested iili Phase One present-
ed more information than any other (risk probability data,[risk comparisons, an
action standard, advice, verbal labels, a risk ladder). Yet| it scored as well as
or better than the other formats on almost all measures of communication
i
j
success, including people's certainty about the risk and their evaluations of the
j. ,
amount of information provided and the helpfulness of the brochure.
"Information overload" may be an issue for still more complex presenta-
tions of risk information, or for audiences that are less interested or less
educated. But no evidence of overload has been found f0r the formats tested
so far. For most uses, in fact, this "maximum information!" condition is proba-
' i -
bly optimal. The likely exceptions would be cases where!no action standard
-------
12 ,
exists or where apathy is a major problem and the communicator wishes to
encourage maximum risk ave'rsion.
The Bottom Line
In general: Don't worry about information overload in written materials, so long
as each piece of information is relatively simple and clearly presented. Always
include an action standard if one exists. Except where maximum risk aversion is .
your'goal, always include risk probability data (if they are available), an appropriately
constructed risk ladder, and advice for different levels. Unless it is misleading,
always include a comparison to normal background if the information is available.
If you are worried about apathy and want to encourage maximum risk aver-
sion: Give people a standard and no other risk information. \~'
If you are .worried about panic and want to encourage mjnimum risk aversion:
Give people advice for different levels, specifying at what levels! you recommend
action and at what levels you recommend doing nothing. Include a risk ladder that
extends to levels higher than those your audience will experience. If the situation
seems likely to generate outrage out of proportion to the risk, include a comparison to
background if at all possible.
If you are trying to help people distinguish high from low/levels of a single
hazard: Construct a risk ladder with the high levels at the top and the low levels at
the bottom. ;
If you are trying to help people distinguish between hazsjrds or understand that
i ' ' '
all levels of a particular hazard are relatively high or low in risk:j Construct a ladder
that is extended upward or downward, so that the hazard you want to depict as low
has all its levels near the bottom of the ladder, and the hazard jyou want to depict as
high has all its levels near the top, !
------- |