Research Report
Tactful or Doubtful?
Expectations of Politeness Explain the Severity Bias in the
Interpretation of Probability Phrases
Jean-Franc¸ois Bonnefon1 and Gae¨lle Villejoubert2
1Laboratoire Travail et Cognition, Centre National de la Recherche Scientifique, Toulouse, France, and
2Leeds University Business School, the University of Leeds, Leeds, United Kingdom
ABSTRACT—When a statement about the occurrence of a
medical condition is qualified by an expression of proba-
bility, such as the word possible, listeners interpret the
probability of the condition as being higher themore severe
the condition. This severity bias can have serious conse-
quences for the well-being of patients. We argue that the
bias is due to a misconception of the pragmatic function
served by the expression of probability. The more severe
the condition, the greater the chance that the listener
construes the expression as a politeness marker rather
than as an uncertainty marker. When this misconception
does not occur, neither should the severity bias. An analysis
of interpretations of probability expressions using a
membership-function approach validates this account. We
discuss the consequences of this bias for the communica-
tion of risk within and outside the medical domain.
Natural language is a poor tool when it comes to communicating
the likelihoods of states of affairs. Numerical probabilities, of
course, are a much better medium for communication of un-
certainty, which is why most people prefer to be given numerical
rather than verbal estimates of likelihoods (Wallsten, Budescu,
Zwick, & Kemp, 1993). Unfortunately, numerical estimates are
often unavailable, forcing people to base their decisions on
phrases as ambiguous as ‘‘X is rather likely’’ or ‘‘Y is highly
possible.’’
Not surprisingly, people are prone to a number of biases in
their interpretation of such expressions of probability. Promi-
nent among these is the severity bias, which is usually demon-
strated with health-related material (Fischer & Jungerman,
1996; Franic & Pathak, 2000; Weber & Hilton, 1990). When a
probability expression (probable, possible, likely, etc.) qualifies a
statement about a patient developing a medical condition or a
side effect of some treatment, listeners interpret the probability
as higher the more severe the condition. The severity bias can
have serious consequences for the well-being of patients.
Overestimating the likelihood of a side effect can encourage
choice of the wrong treatment. Misunderstanding the likelihood
of developing a condition can make communication between
doctor and patient frustrating and counterproductive. This po-
tential for damage is further increased by society-wide policies
such as the European recommendation to give verbal estimates
of the likelihood of side effects (European Commission, 1998).
Some experts have asked for the suspension of such policies
until research has yielded further insight into the interpretation
of probability expressions (Berry, Raynor, & Knapp, 2003).
However, only meager understanding has been achieved so far.
Insight into the mechanics of the severity bias has gone no
further than Weber and Hilton’s (1990) initial explanation that
‘‘more severe events may draw attention to potentially higher
probability levels, something that might be labeled a ‘worry
effect’ ’’ (p. 788). We propose that the severity bias derives from
what Goffman (1967) and Brown and Levinson (1978/1987)
have identified as a fundamental mechanism of human social
life, face-work. All humans project a sense of positive identity
and public self-esteem called ‘‘face’’ and are motivated to sup-
port their own and other people’s face in social interactions.
Many actions, called face-threatening acts, can induce a loss
of face (e.g., disagreeing with, criticizing, giving orders to, or
embarrassing other people). When such an action is performed,
the actor is likely to resort to one of many linguistic strategies
that mitigate the face threat. Among these strategies is the use of
probability expressions, not to communicate degrees of uncer-
tainty, but rather to reduce the impact of face-threatening acts.
Our explanation of the severity bias involves two assumptions.
The first assumption is that in addition to communicating like-
lihood, which is their standard function, probability expressions
can be used as face-management devices when they qualify
Address correspondence to Jean-Franc¸ois Bonnefon, Laboratoire
Travail et Cognition, Maison de la Recherche, 5 Alle´es A. Machado,
31058 Toulouse Cedex 9, France, e-mail: bonnefon@univ-tlse2.fr.
PSYCHOLOGICAL SCIENCE
Volume 17—Number 9 747Copyrightr 2006 Association for Psychological Science

face-threatening acts (e.g., ‘‘you might be misinformed,’’ ‘‘it is
possible you will have to pay for my lunch’’). There is ample
evidence for this claim, both qualitative (Brown & Levinson,
1978/1987) and quantitative. For example, Youmans (2001)
reported a rate of 787 probability terms per 20,000 words (from
American English speakers), and in 39% of these cases, the
probability expression was used for face-management purposes
rather than to communicate likelihood. The second assumption
is that probability expressions imply high probabilities when
they are used for face management. The word possibly in ‘‘it is
possibly going to snow tomorrow’’ has the function of commu-
nicating likelihood, and accordingly denotes a moderate prob-
ability of snowfall. However, in ‘‘your bad breath is possibly the
reason people shun you,’’ possibly has nothing to do with un-
certainty, and everything to do with face management. In this
context, it denotes a high likelihood. We obtained preliminary
evidence for this claim in a previous study (Bonnefon & Ville-
joubert, 2005), in which we found that the terms possibly and
probably denoted higher likelihoods when they qualified criti-
cisms or impositions than when they qualified non-face-threat-
ening contents.
Probability expressions denote the same probability whatever
the condition they qualify, as long as they are perceived to
perform their likelihood-communication function and as long as
all other things are equal (in particular, that the conditions have
the same base rate). But when a physician tells a patient that he
or she is going to develop a medical condition, the patient’s
social face is threatened, and the threat is greatest when the
condition is most severe. Therefore, the more severe the patient’s
condition, the more likely a probability expression will be in-
terpreted as a face-management device, rather than as a like-
lihood-communication device. Increasingly severe conditions
increase the number of speakers and hearers who switch to a
face-management interpretation of probability expressions. This
shift in interpretation, in turn, increases the average probability
attached to the expression—hence the severity bias.
METHOD
Participants
Participants were recruited by third-year psychology students at
the University of Toulouse, France, as a course requirement.
Each student made a list of several men and women who were
older than 18 and not studying psychology, randomly drew one
man and one woman from his or her list, and asked them to take
part in the study. Of the 810 participants in the final sample (401
men, 409 women; mean age 5 31.2, SD 5 12.8), 21% had
completed graduate school, 47% had an undergraduate educa-
tion, 19% had graduated from high school only, and the re-
maining 13% had not graduated from high school. The sample
included a large proportion of students (39%), but 61% came
from a great variety of non-student professions (including 8%
who were unemployed).
Materials and Procedure
Data collection focused on the word possibly. The numerical
interpretation of this word was assessed by eliciting its fuzzy
membership function (Zadeh, 1965). This membership function
assigns to each value of the probability line [0,1] a number that
represents the degree of membership of that value in the concept
defined by the phrase. Degree of membership is usually ex-
pressed as a real number from 0 to 1, such that memberships of
0 denote probabilities that are absolutely not in the concept,
and memberships of 1 denote probabilities that are perfect
exemplars of the concept. Other values represent intermediate
degrees of membership. Membership functions (originally sug-
gested by Wallsten, Budescu, Rapoport, Zwick, & Forsyth,
1986, and Rapoport, Wallsten, & Cox, 1987) provide subtle and
rich representations of the meaning of probability expressions,
and the use of such functions has been carefully validated in
many studies (for reviews, see Budescu & Wallsten, 1995, and
Karelitz & Budescu, 2004).
Membership functions were elicited using the multiple-
stimuli method introduced in Budescu, Karelitz, and Wallsten
(2003). Participants were asked to imagine that their family
doctor had announced they would ‘‘possibly’’ develop each of
two medical conditions during the year to come. One condition
was deafness, and the other was insomnia (order of presentation
was counterbalanced). Insomnia and deafness are of similar
prevalence in the French population, from which participants
were sampled (i.e., both had incidence rates of about 4% in
2000). After having read, ‘‘The doctor tells you, you will possibly
suffer from insomnia [deafness] soon,’’ participants were asked:
‘‘Does the doctor think the probability that you will suffer from
insomnia [deafness] soon is . . . ?’’ This question was followed by
the values 10%, 20%, 30%, and so on, up to 100%. Participants
provided a judgment for each of the 10 percentages, using a 10-
point scale anchored at absolutely not and absolutely. The task
was then repeated with the second medical condition (deafness
or insomnia). In addition, participants were asked to say whether
developing deafness or developing insomnia was worse news.
Individuals who failed to answer (n 5 16) or who judged in-
somnia to be worse news than deafness (n 5 131) were filtered
out, yielding the final sample of 810 participants. This choice
substantially improved the clarity of the analyses.
Finally, participants were asked whether the doctor was
qualifying deafness as ‘‘possible’’ because (a) he was not sure it
would happen or (b) he wished to announce the news tactfully.
The same question was asked with respect to insomnia, to assess
in each case which speech function participants believed the
probability term was intended to perform.
RESULTS
‘‘Possible deafness’’ was judged more probable than ‘‘possible
insomnia.’’ Membership functions were computed by averaging
membership judgments across participants. The top panel of
748 Volume 17—Number 9
Politeness and Verbal Uncertainty

Figure 1 depicts these membership functions overall, without
distinguishing which speech function participants assigned to
possible. These functions show the expected severity bias, as the
function for deafness peaks at a higher probability than the
function for insomnia. The bias was confirmed by computing the
two functions’ peaks (see Table 1). The peak for each function
was calculated by averaging for each participant the probability
values that received the highest membership ratings and then
averaging the values obtained across participants.1 The peak of
the function for deafness was higher than the peak for insomnia,
t(815) 5 7.90, prep > .999 (Killeen, 2005), d 5 0.23 (the 95%
confidence interval for this difference was .06–.09).
It appears that one is uncertain about ‘‘possible insomnia,’’
but tactful about ‘‘possible deafness.’’ As indicated in the last
column of Table 1, the proportion of participants who judged that
the doctor used the term possible for face-management purposes
was much greater for deafness (60%) than insomnia (17.4%; z5
17.1, prep> .999, Cohen’s h5 0.9). One noteworthy result is that
only 29 participants judged that the doctor was tactful about
insomnia but uncertain about deafness. Thus, less than 4% of
the participants directly contradicted our hypothesis that more
severe conditions increase the likelihood that probability ex-
pressions will be perceived as tactful.
The numerical interpretation of possible is a function of its
being tactful or uncertain. The lower-left panel of Figure 1 de-
picts the membership functions of possible for insomnia and for
deafness only for participants who judged that this word com-
municated likelihood. In contrast, the lower-right panel depicts
the same function for participants who judged the term to be
tactful, serving face-management motives. The functions within
each panel are similar, especially in the case of the lower-left
panel (likelihood communication), but the difference between
the two panels is dramatic, with the functions in the right panel
peaking at higher probabilities than those in the left panel. With
respect to insomnia, the peak of the face-management mem-
bership function is greater than the peak of the likelihood-
communication function (see Table 1 for descriptive statistics),
t(808)5 4.9, prep> .999, d5 0.46 (the 95% confidence interval
for this difference was .09–.21). Similarly, with respect to
deafness, the peak of the face-management membership func-
tion is greater than that of the likelihood-communication func-
tion, t(808)5 8.5, prep> .999, d5 0.61 (with a 95% confidence
interval of .14–.23).
Finally, we consider the subsample of participants (n5 295)
who judged that the doctor was expressing uncertainty about
both insomnia and deafness. If politeness expectations do un-
derlie the severity bias, these participants should not have
manifested any bias in their numerical interpretations. Indeed,
the peaks of the membership functions for insomnia (M 5 .57,
SD5 .35) and deafness (M5 .56, SD5 .34) were practically the
same for these participants, t(294) 5 0.8, prep 5 .55, d 5 0.05
(the 95% confidence interval for the difference between peaks
was
.01–1.03). The two peaks were strongly correlated, r5 .87.
DISCUSSION
Previous research has established that probability expressions
not only communicate degrees of uncertainty, but also serve
other pragmatic functions, such as expressing perspective and
Fig. 1.Membership functions of ‘‘possible’’ deafness and insomnia, for the
whole sample (top panel, n 5 810), for those participants who under-
stood ‘‘possible’’ as serving likelihood-communication purposes (lower-left
panel, n5 669 for insomnia and n 5 324 for deafness), and for those par-
ticipants who understood ‘‘possible’’ as serving face-management purposes
(lower-right panel, n 5 141 for insomnia and n 5 486 for deafness).
TABLE 1
Mean Peaks of the Membership Functions Attached to ‘‘Possible’’
Insomnia and Deafness, as a Function of the Understood
Communicative Function of the Probability Term
Medical condition and
communicative function M (SD) n
Insomnia
Likelihood communication .57 (.33) 669
Face management .71 (.30) 141
Overall .59 (.33) 810
Deafness
Likelihood communication .55 (.34) 324
Face management .74 (.28) 486
Overall .67 (.32) 810
1It is common for there to be a small discrepancy between the numerical
computation of the peak and its graphical representation. Accordingly, the
values in Table 1 do not exactly match the peaks of the functions in Figure 1.
Volume 17—Number 9 749
Jean-Franc¸ois Bonnefon and Gae¨lle Villejoubert

drawing attention to the occurrence or the nonoccurrence of the
event to which they refer (Sanford & Moxey, 2003; Teigen &
Brun, 2003). They may also serve social face-management
purposes. We conjectured that the severity bias in interpreta-
tions of verbal probability expressions results from the use of
such expressions as face-management devices whose function
is to safeguard the feelings of people who are receiving face-
threatening news.
We tested this conjecture in the medical context of a doctor
telling a patient that he or she might develop a mild or severe
medical condition. We expected that the face-management in-
terpretation would be more likely when the condition was se-
vere, and that a face-management interpretation would lead to
an overestimation of the probability of this condition. Results
replicated the severity-bias effect. The same probability word
(possibly) was judged to communicate higher numerical proba-
bilities when it qualified a more severe condition (deafness) than
when it qualified a less severe but equally prevalent condition
(insomnia). Furthermore, when the probability term qualified
the more severe condition, most participants thought it served a
face-management purpose. Also, those who believed the term
was used as a face-management device thought that the condi-
tion it qualified was substantially more likely to occur than did
those who thought the term was communicating a vague likeli-
hood. But participants who believed the probability term was
intended to communicate the likelihood of occurrence for both
diseases did not exhibit the severity bias.
Thus, people recognize that the more severe a condition is, the
more threatening is the news that one has this condition. They
also understand that it is polite and tactful to mitigate such face-
threatening news by using linguistic moderators, such as prob-
ability terms. Finally, they recognize that a probability term used
as a face-management device does not refer to the probability of
the event it qualifies.
Although our experimental test of this account was limited to
the word possible, the account should apply to other expressions
of probability as well. However, some terms may be less ap-
propriate for expressing politeness than others are. For example,
we suspect that probable is a less plausible politeness term than
possible. If so, fewer respondents would interpret information
communicated with probable as being intended to be tactful. But
this would not undermine our main findings. Although a phrase
with the term probable in it is less likely to be interpreted as
serving a face-management purpose than is a phrase with the
term possible, a statement using probable would still be judged to
indicate a higher numerical probability when it is interpreted as
tactful than when it is interpreted as communicating likelihood
information directly (Bonnefon & Villejoubert, 2005). Moreover,
when probable is interpreted as communicating likelihood, it
should receive the same numerical interpretation whatever the
severity of the condition to which it refers.
A misunderstanding about which function a probability
phrase is intended to serve could lead to a discrepancy between
the level of probability a doctor intends to communicate and the
level of probability understood by the patient. If the patient
interprets a particular phrase as a face-management device but
the doctor used it to communicate a vague likelihood, the patient
may overestimate the probability the doctor intended. Thus, this
research suggests that measures should be taken to ensure that
speakers and hearers assign the same communicative function
(likelihood communication or face management) to a given
probability phrase to improve risk communication.
Acknowledgments—We thank Reid Hastie, Mansur Lalljee,
Linda Moxey, and Karl Teigen.
REFERENCES
Berry, D., Raynor, D., & Knapp, P. (2003). Communicating risk of
medication side effects: An empirical evaluation of EU recom-
mended terminology. Psychology, Health and Medicine, 8, 251–
263.
Bonnefon, J.F., & Villejoubert, G. (2005). Communicating likelihood
and managing face: Can we say it is probable when we know it to be
certain? In B.G. Bara, L. Barsalou, & M. Buciarelli (Eds.), Pro-
ceedings of the 27th annual conference of the Cognitive Science
Society (pp. 316–321). Mahwah, NJ: Erlbaum.
Brown, P., & Levinson, S.C. (1987). Politeness: Some universals in
language usage. Cambridge, England: Cambridge University
Press. (Original work published 1978)
Budescu, D.V., Karelitz, T.M., & Wallsten, T.S. (2003). Determining the
directionality of probability words from their membership func-
tions. Journal of Behavioral Decision Making, 16, 159–180.
Budescu, D.V., & Wallsten, T.S. (1995). Processing linguistic proba-
bilities: General principles and empirical evidence. In J. Buse-
meyer, D.L. Medin, & R. Hastie (Eds.), Decision making from a
cognitive perspective (pp. 275–318). San Diego, CA: Academic
Press.
European Commission. (1998). A guideline on the readability of the
label and package leaflet of medicinal products for human use.
Brussels, Belgium: Pharmaceuticals Committee, European Com-
mission.
Fischer, K., & Jungerman, H. (1996). Rarely occurring headaches and
rarely occurring blindness: Is rarely rarely? Journal of Behavioral
Decision Making, 9, 153–172.
Franic, D.M., & Pathak, D.S. (2000). Communicating the frequency of
adverse drug reaction to female patients. Drug Information Jour-
nal, 34, 251–272.
Goffman, E. (1967). Interaction ritual: Essays on face-to-face behavior.
Garden City, NY: Anchor Books.
Karelitz, T.M., & Budescu, D.V. (2004). You say ‘‘probable’’ and I say
‘‘likely’’: Improving interpersonal communication with verbal
probability phrases. Journal of Experimental Psychology: Applied,
10, 25–41.
Killeen, P.R. (2005). An alternative to null-hypothesis significance
tests. Psychological Science, 16, 345–353.
Rapoport, A., Wallsten, T.S., & Cox, J.A. (1987). Direct and indirect
scaling of membership functions of probability phrases. Mathe-
matical Modeling, 9, 397–417.
Sanford, A.J., & Moxey, L.M. (2003). New perspectives on the ex-
pression of quantity. Current Directions in Psychological Science,
12, 240–243.
750 Volume 17—Number 9
Politeness and Verbal Uncertainty

Teigen, K.H., & Brun, W. (2003). Verbal expressions of uncertainty and
probability. In D. Hardman & L. Macchi (Eds.), Thinking: Psy-
chological perspectives on reasoning, judgement, and decision-
making (pp. 125–145). Chichester, England: Wiley.
Wallsten, T.S., Budescu, D.V., Rapoport, A., Zwick, R., & Forsyth, B.
(1986). Measuring the vague meanings of probability terms.
Journal of Experimental Psychology: General, 115, 348–365.
Wallsten, T.S., Budescu, D.V., Zwick, R., & Kemp, S.M. (1993). Pref-
erence and reasons for communicating probabilistic information
in numerical or verbal terms. Bulletin of the Psychonomic Society,
31, 135–138.
Weber, E.U., & Hilton, D.J. (1990). Contextual effects in the inter-
pretation of probability words: Perceived base-rate and severity of
events. Journal of Experimental Psychology: Human Perception
and Performance, 16, 781–789.
Youmans, M. (2001). Cross-cultural differences in polite epistemic
modal use in American English. Journal of Multilingual and
Multicultural Development, 22, 57–73.
Zadeh, L.A. (1965). Fuzzy sets. Information and Control, 8, 338–353.
(RECEIVED 9/14/05; ACCEPTED 12/14/05;
FINAL MATERIALS RECEIVED 1/12/06)
Volume 17—Number 9 751
Jean-Franc¸ois Bonnefon and Gae¨lle Villejoubert