Pain and negative emotions in the face:
judgements by health care professionals
Judith Kappessera
, Amanda C. de C. Williamsb,*
a
Department of Psychology, University of Bath, Claverton Down, Bath BA2 7AY, UK
b
Department of Psychological Medicine, GKT Medical School, University of London, St Thomas' Hospital, London SE1 7 EH, UK
Received 23 October 2001; received in revised form 27 February 2002; accepted 25 March 2002
Abstract
Facial expression of pain has rarely been researched in the context of facial expression of negative emotions with which it may occur. The
main aim of the study was to investigate how pain expression resembled or differed from that of other negative emotions (fear, anger,
sadness, surprise, disgust and embarrassment), using multidimensional scaling, a dimensional approach to understanding relationships
among emotions. As possible misidentification of facial expressions by participants could distort those results, a judgement study as a
categorical approach was conducted to examine the accuracy of identification of pain and negative emotion facial expressions. The sample
was health care professionals. Identification of pain was good (unbiased hit rate 58.8%), but less than all other negative emotions. Confidence
in ratings approximated accuracy of identification. Multidimensional scaling revealed two dimensions: the first distinguished embarrassment
from all other emotion expressions; the second separated pain, sadness and anger from fear, surprise and disgust. Possible explanations for
these findings were sought in patterns of facial action units, and in the messages conveyed by the expressions according to Fridlunďs
Behavioural Ecology View. q 2002 International Association for the Study of Pain. Published by Elsevier Science B.V. All rights reserved.
Keywords: Facial expression; Pain; Pain behaviour
1. Introduction
Investigation of facial expressions has a long tradition in
emotion research. In the last 50 years, research in facial
expression of emotion has followed two particular lines of
investigation (Wagner, 1997): component studies, in which
muscle actions constituting the expression are identified and
quantified, usually by the facial action coding system
(FACS: Ekman and Friesen, 1978), and judgement studies,
which address the information conveyed by the global
expression. The latter provide strong evidence of at least
five discrete expressions of emotion (happiness, fear,
anger, sadness and disgust) which are universally recognised,
and to which many would add surprise and contempt
(Ekman, 1992; Ekman and Friesen, 1986). However, the
exact number of distinct emotional expressions is still undetermined.
Keltner and colleagues, for example, initiated
research concerning the self-conscious emotions and
found evidence for the distinctiveness of the facial expression
of embarrassment (Keltner and Buswell, 1996). In
contrast to the categorical approaches to emotional prototypes
some researchers describe dimensional models of
emotion such as Russell (Carroll and Russell, 1996).
However, studies by Young et al. (1997) support a categorical
rather than a dimensional account of emotions.
Although many emotion theorists assume pain not to be
an emotion and have only rarely included pain facial expressions,
there are good reasons why the methodologies of
these studies provide appropriate tools for investigation of
the pain face and why the pain face should be considered in
connection with established facial expressions of emotions.
The emotional quality of pain is emphasised in the widely
used definition of pain (International Association for the
Study of Pain, 1979). Furthermore, viewed from an evolutionary
perspective, facial expressions signal emotional
experience (Prkachin, 1997). This is further supported by
the `Behavioural Ecology View' of faces proposed by
Fridlund (1994, 1997), a complementary approach to facial
expressions derived from modern accounts of the genetic
and cultural evolution of signalling behaviour. The Behavioural
Ecology View of faces constitutes a model in its
view of how facial expressions evolved, what they signify
and how they function in our everyday lives, in contrast with
Pain 99 (2002) 197­206
0304-3959/02/$20.00 q 2002 International Association for the Study of Pain. Published by Elsevier Science B.V. All rights reserved.
PII: S0304-3959(02)00101-X
www.elsevier.com/locate/pain
* Corresponding author. INPUT Pain Management Unit, St. Thomas'
Hospital, London SE1 7EH, UK. Tel.: 144-207-922-8107; fax: 144-
207-922-8229.
E-mail address: amanda.williams@kcl.ac.uk (A.C. de C. Williams).
research emphasising the centrality of emotions in explaining
facial movements. The latter may be described as
`Emotions View', in which facial movements are directly
linked to emotions and facial expressions of emotions are
reflex-like readouts of those emotions (Buck, 1994).
Fridlund, on the contrary, regards facial expressions as
social tools which are `reaď according to the context of
the interaction in which they occur. As he assumes that
faces exert their influence in the particular context of their
occurrence, they may only be interpreted within this
context. For instance, rather than describe a face as `saď,
Fridlund would say that this face signals the attempt to
recruit help and care. Accordingly, a pain face would signal
suffering and attempt to gain relief from pain and emotional
suffering.
The facial expression of pain has been investigated not in
the context of emotion, but as a class of `pain behaviour'
(Craig, 1980; Fordyce, 1976), other classes of which include
verbal communication of pain, paralinguistic vocalisations,
distinct movements or postures, and visible physiological
changes (Craig, 1992; Craig and Prkachin, 1983; Craig et
al., 2001). Facial expression is, socially, the most prominent
of these (Craig et al., 2001; Prkachin et al., 1983; von
Baeyer et al., 1984).
Component study methodology has identified specific
facial movements associated with pain, including lowering
the brow, narrowing the eyes by tightening the lids and raising
the cheeks or even fully closing the eyes, raising the upper
lip, deepening the nasolabial fold and wrinkling the nose as
well as opening the lips and mouth in varying degrees. This is
consistent across a range of experimental pain modalities
(Craig and Patrick, 1985; Galin and Thorn, 1993; LeResche
et al., 1992; Patrick et al., 1986; Prkachin, 1992), and across
different clinical pain conditions (Craig et al., 1991; Hadjistavropoulos
and Craig, 1994; LeResche, 1982; LeResche and
Dworkin, 1988; Prkachin and Mercer, 1989).
Using the FACS, the expression of pain can be differentiated
from the standard emotion expressions, with some of
which it shares certain facial action units as displayed in
Table 1 (Craig, 1992; LeResche, 1982). Judgements by
observers show identification of pain expression to be well
above chance level (Keltner and Buswell, 1996; LeResche
and Dworkin, 1984); further, it can be distinguished from
other emotions in photos (Boucher, 1969; Haidt and Keltner,
1999; Keltner and Buswell, 1996), and other emotions
are seldom mistaken for pain when it is not a stimulus but
provided as a response option (e.g. Carroll and Russell,
1996). However, the pain face has also found to be blended
with other facial expressions of emotions such as disgust,
contempt, anger, fear, and sadness (LeResche, 1982;
LeResche and Dworkin, 1988). In a study by Hale and
Hadjistavropoulos (1997) patients undergoing a routine
blood test were videotaped. Not only facial expression of
pain but also facial expressions of disgust, anger, fear, and
happiness varied significantly across the conditions (baseline,
swabbing, and venepuncture).
One focus of the study of pain expression has been observers'
underestimation of pain by reference to the sufferer's
evaluation. In a study by Prkachin et al. (1994), observers
estimated the amount of pain experienced by patients with
shoulder injuries, from the patients' faces seen on videotape:
compared with patients' own ratings, facial actions
coded with the FACS provided a sensitive measure of
pain whereas observers' judgements systematically underestimated
patients' pain by as much as 80%. While direct
facial measurement provides a more sensitive measure of
pain than observers' judgements, attempts to enhance observers'
sensitivity to facial expression of pain by training (e.g.
Galin and Thorn, 1993; Solomon et al., 1997) have been
disappointing. Medical and paramedical professionals in
particular appear to show an underestimation bias in the
course of clinical work (Choiniere et al., 1990; Teske et
al., 1983; Zalon, 1993), although accurate assessment of
pain is a prerequisite for adequate treatment.
Facial expression of pain, therefore, has been established
in terms of facial action units in relation to other emotions,
but judgement studies of pain expression have focused on
authenticity or estimation of pain, not (apart from in the
embarrassment studies of Keltner and colleagues) the accuracy
of distinguishing pain from other emotions. The
present study addressed three questions concerning facial
expression of pain and other negative emotions as perceived
by health care professionals. Our main aim was to identify
the dimensions on which pain expression resembled or
differed from that of other negative emotions (fear, anger,
sadness, surprise, disgust, embarrassment), using multidimensional
scaling (MDS) of similarity­dissimilarity
comparisons. Possible interpretations for similarities and
differences drew on the Emotions View and Fridlunďs
Behavioural Ecology View. As possible misidentification
of facial expressions by participants could distort the results
of the similarity­dissimilarity comparisons, a judgement
study was conducted to examine the question: to what extent
each of the standard facial expressions could be identified
correctly by participants. The results of this judgement
study also addressed the third question of the extent to
which underestimation of pain by health care professionals
might be due to their not recognising pain faces, due to their
confusing pain with other facial expressions of emotion and/
or due to their being less confident in identifying pain,
although it could not constitute a direct comparison of
these mechanisms.
2. Method
2.1. Participants
Medical and nursing staff working for Accident and
Emergency (ER) Departments in two London hospitals
were asked to volunteer as participants, since they are
exposed to pain faces daily. They were approached before,
J. Kappesser, A.C. de C. Williams / Pain 99 (2002) 197­206198
J.Kappesser,A.C.deC.Williams/Pain99(2002)197­206199
Table 1
Prototypical AUs which occur with sadness, fear, anger, disgust, surprise, embarrassment and paina
Emotion Au
1 2 4 5 6 7 9 10 11 12 15 16 17 20 22 23 24 25 26 27 43 51 54 64
Inner
brow
raiser
Outer brow
raiser
Brow
lower
Upper
lid raiser
Cheek
raiser
Lid
tighten
Nose
wrinkler
Upper
lip raiser
Nasol.
furrow
deepen
Lip
corner
puller
Lip
corner
depr.
Lower
lip depr
Chin
raiser
Lip
stretch
Lip
funnel
Lip
tighten
Lip
pressor
Lips
part
Jaw
drop
Mouth
stretch
Eyes
closed
Head
turn left
Head
down
Eyes
down
Sadness B B A A B A A A A
Fear B B B B A B B B
Anger B B B A A B B A B B
Disgust B B A A A
Surprise B B B A A
Embarrassment B B B B B
Pain B B B B B A A A A A B
a
The prototypical pain face is mainly based on the results by Prkachin (1992), but considers some extra AUs which are discussed in the literature. B, AU which characteristically occurs with emotion
prototypes; A, may occur with prototype and/or variant; __, AUs which occur together.
during or after their work shift; aim and content of the study
were explained and they were asked for their consent.
2.2. Photos
The photos which were used for this study were kindly
provided by Dacher Keltner, and are one of the three posers
of those used in Keltner and Buswell (1996) and Haidt and
Keltner (1999). The photos were generated on the basis of
emotion prototypes using the FACS, and include pain alongside
various emotions. The coloured photos show a male
face, neck and part of the shoulders against a coloured background.
The individual who served as a model produced the
facial expressions according to the facial action units specified
by Ekman and Friesen (1978). While the use of only one
model risks confounding variance due to model with
variance due to emotion expressions, it was thought more
important to avoid the use of different models for different
expressions of emotion. Time constraints precluded the
presentation of multiple models across all expressions of
emotion.
Sadness, fear, anger, surprise, disgust and embarrassment
were chosen as additional negative expressions to pain. The
photos of the facial expressions of anger, fear, sadness,
disgust and surprise are based on their prototypical display
according to Ekman and Friesen (1978). The facial expression
of embarrassment is based on the findings of Keltner
(1995) and the one for pain on results by Patrick et al.
(1986).
2.3. Procedure
After agreeing to participate in the study, participants'
age, sex, qualification, overall length of work experience
and length of work experience in the ER were recorded.
Participants were asked to imagine that the photo represented
a patient attending the ER. In the first part of the
study, participants were asked to compare each possible
pair of the seven photos regarding its similarity or dissimilarity
with pain. In the second part, participants had to
identify the facial emotion shown on every photo. The
second part, a judgement study, is effectively a validity
check; a necessary prerequisite which allows for analysing
the similarity­dissimilarity comparisons.
2.3.1. Procedure for the similarity­dissimilarity
comparisons
Participants were presented with numerical rating scales
which consisted of 11 points from 0, `exactly the same', to
10, `completely differenť. Participants were informed that
they would view pairs of photos and that they should rate the
extent of similarity/difference between the pair in terms of
acute physical pain, that is, that two faces could be judged
similar in expressing pain or in not expressing pain, and that
where they differed their dissimilarity was judged in terms
of degree of closeness to pain. Since these similarity­
dissimilarity comparisons assume an internal scale along
which the judgements are made, participants previewed
the set of seven photographs (sadness, fear, anger, surprise,
disgust, embarrassment and pain) for 15 s. After previewing
the seven photos, participants were asked to compare pairs
of photos. Twenty-one pairs resulted from all pairings of the
seven faces, and the order of presentation was randomised
across participants. A second presentation of one of the 21
pairs was added in order to assess the stability of the ratings,
but to avoid repeating of the same pair's numerical rating
scale on the same page, a semi-random order rather than a
random one was used.
2.3.2. Procedure for the judgement study
The order of the seven photos was randomised by a
random number table. Then the photos were presented to
participants serially for each of the participants in random
order. All emotion terms (sadness, fear, anger, surprise,
disgust, embarrassment), plus pain, were offered with each
photo and participants were asked to select the best description
of the face, or the option of none, to which participants
could append their own best descriptive term. Additionally,
they were asked to rate their confidence in each rating on a
numerical scale which consisted of 11 points from 0, `none',
to 10, `totaľ.
After participants completed the second task, they were
offered the chance to ask questions concerning content and
aim of the whole study.
2.4. Analysis of data
2.4.1. Analysis of the similarity­dissimilarity comparisons
To estimate the stability of the similarity­dissimilarity
ratings, bivariate correlations and differences in the ratings
between the repeated pairs were calculated. The data were
then subjected to a MDS procedure (SPSS implementation
of ALSCAL) to provide spatial representation of similarities,
that is, grouping by latent structure (Schiffman et al.,
1981).
2.4.2. Analysis of the judgement study
The judgement study examines whether the emotions
displayed on the photos are identified as such. Of particular
interest was how accurately pain is identified, with what
other emotions pain is confused, what other emotions are
misread for pain and how confident participants felt about
identifying pain.
The results of the judgement study were summarised in a
confusion matrix (see Table 2 for example). If the focus is
the accuracy of the response, one measure is the simple hit
rate, i.e. the proportion of correctly identified target stimuli.
In Table 2 this is calculated by a=a 1 b (equivalent of
positive predictive value of a diagnostic test). Wagner
(1993, 1997) recommends tranforming this to an unbiased
hit rate, which takes into account misidentification of the
target (cell c in Table 2). The unbiased hit rate (HU) is the
product of two conditional probabilities, that a target stimuJ.
Kappesser, A.C. de C. Williams / Pain 99 (2002) 197­206200
lus will be correctly identified a=a 1 b, designated H1),
and that the response to a stimulus will be correct
a=a 1 c, equivalent to sensitivity of a diagnostic test,
designated H2). As for simple hit rates, values for unbiased
hit rates lie within the range of 0­1.
For the confidence ratings, the means and standard deviations
were calculated and differences investigated according
to correct or incorrect identification.
3. Results
3.1. Description of the sample
Altogether 60 nurses and doctors of two ER Departments
took part in the study. Among the participants were more
women (68.3%) than men and more nurses (73.3%) than
doctors. Whereas the distribution of nurses and doctors
among the male participants was nearly equal (47.4%
male nurses), there were more than 5.5 times as many nurses
(85.4%) as doctors among the female volunteers.
All the nurses and doctors were qualified except one final
year medical student. The mean age of participants was 28.5
years (SD 4.83 years). Participants had a median 3.25 years
of experience as a nurse or doctor (range 0­26 years) and a
median 2.0 years (range 0­15 years) of working in ER.
As the judgement study is a necessary prerequisite which
allows for analysing the similarity­dissimilarity comparisons
(possible misidentification of facial expressions by
participants could distort the results of the similarity­
dissimilarity comparisons), we show those results first.
The order was reversed in the study in order to avoid participants'
judgement processes affecting their responses to the
MDS.
3.2. Results of the judgement study
As a validity check for the similarity­dissimilarity
comparisons, the judgement study examines whether the
emotions displayed on the photos are identified as such.
Of additional interest was how accurately pain is identified,
with what other emotions pain is confused, what other
emotions are misread for pain and how confident participants
felt about identifying pain.
3.2.1. Confusion matrix and hit rates
The results of the judgement study ­ the confusion matrix
and the hit rates ­ are displayed in Tables 3 and 4. Recognition
for all facial expressions is far higher than chance level
(1/8  12.5% or 7.5 participants choosing this expression),
irrespective of whether the simple hit rate or the unbiased hit
rate was used. One-tailed binomial tests for the simple hit
rates reveal P-values ,0.001 for all facial expressions.
Seventy percent of participants (42 of 60) identified the
pain face as showing pain; of the remaining 30%, disgust
was chosen by 11 (18.3%, for which a one-tailed binomial
test was not significant, P  0:121 if the undecided participanťs
judgement is assigned to disgust, P  0:059 if it is
assigned to pain), embarrassment by three (5%) and fear by
one (1.67%). Three participants (5%) responded that none
J. Kappesser, A.C. de C. Williams / Pain 99 (2002) 197­206 201
Table 3
Confusion matrix of emotion category responses to photo stimuli
Photo Response Sum
Sadness Fear Anger Surprise Disgust Embarrassment Pain None of them
Sadness 58 ­ ­ ­ ­ ­ ­ 2 60
Fear 1 42 2 10 1 2 2 ­ 60
Anger 1 1 50 ­ 2 ­ 4 2 60
Surprise ­ 3 ­ 57 ­ ­ ­ ­ 60
Disgust ­ ­ 5 ­ 51 ­ 2 2 60
Embarrassment 1 1 ­ ­ ­ 52 ­ 6 60
Pain ­ 1 ­ ­ 11 3 42 3 60
Sum 61 48 57 67 65 57 50 15 420
Table 2
Example of confusion matrix
Stimulus Response Sum
Response 1 Response 2
Stimulus 1 a b a 1 b
Stimulus 2 c d c 1 d
Sum a 1 c b 1 d a 1 b 1 c 1 d
Table 4
Simple hit rates (H1, H2) and unbiased hit rate (HU) for the seven photos
Photo Hit rates
H1 (%) H2 (%) HU (%)
Chance 1/8  12.5% 1/7  14.3%
Sadness 96.7 95.1 91.9
Surprise 95.0 85.1 80.8
Embarrassment 86.7 91.2 79.1
Anger 83.3 81.7 73.1
Disgust 85.0 78.5 66.7
Fear 70.0 87.5 61.3
Pain 70.0 84.0 58.8
of the emotion terms offered could describe the pain face:
two offered no alternative, and a third could not decide
between pain and disgust. None of the participants thought
that the pain face would be best described by the terms
sadness, anger or surprise. Furthermore, the confusion
matrix shows that pain as a response category was chosen
50 times altogether: 42 (84%) to the pain face, 4 (8%) to the
anger face, and two times (4%) to each of the fear and the
disgust faces. None of the participants found that the sad,
surprised or embarrassed photo would be best described as
pain.
3.2.2. Confidences
Although recognition for the pain face (HU  58:80%;
H1  70%) is far higher than chance level, both hit rates
are the lowest when compared to the other facial expressions.
This resembles the rank order of confidence ratings in
Table 5. The scale for the confidence ratings ranged from 0
(`none') and 10 (`totaľ), but the lowest point used by
respondents was 2, and medians ranged from 7 to 9 (Table
5). Since distributions were skewed, medians were used for
comparisons.
The Pearson's correlation coefficient between the median
of the confidence ratings and the unbiased hit rates is 0.81
P  0:027. However, there were significant differences
between confidence ratings (Friedman x2
 43:844;
P , 0:0005). Post hoc analysis, using a Wilcoxon signed
rank test and a more stringent P-value (0.05/20  0.0025),
indicated that the differences were due to significantly
higher confidence ratings for the sadness face than for
each other face (z from 3.165 to 4.781, P  0:002 to
,0.001), and significantly lower ratings for the pain face
when compared with anger, embarrassment and surprise (z
from 3.045 to 3.464, P  0:002 to 0.001).
So far, no distinction has been made between the confidence
of subjects who identified a facial expression
correctly and subjects who did not. Fig. 1 shows the confidence
ratings for both groups. Provided that there were more
than seven subjects in the smaller (always incorrect) group,
a Mann­Whitney test was used to test for median confidence
difference between groups. The two groups for
which the hit rate was lowest, fear and pain, also had the
largest differences in median confidence rating, but only the
difference for the fear photo attained statistical significance
(z  2:147, P  0:032).
3.3. Results of the similarity­dissimilarity comparisons
3.3.1. Reliability check of ratings
The differences in ratings (210 to 110) for two presentations
of the same pair form a normal distribution with a mean
of 10.57 and a standard deviation of 2.79. Pearson's correlation
coefficient between the repeated pairs is 0.595. For 25%
of ratings, there was perfect agreement and for more than
50% of the ratings, the difference was not greater than 1
point on the rating scale. To ensure stability of the ratings,
participants whose responses differed by approximately two
standard deviations (differences $ 5) were excluded from
the analysis; this excluded seven participants and the correlation
coefficient increased from 0.595 to 0.815.
3.3.2. Identifying the stimulus space
The dimensionality of a stimulus space is determined both
by maximum increase in the averaged squared correlations
(RSQ) and by the corresponding decrease in stress, to the
extent that they covary inversely. Should they increase or
decrease together, the RSQ is the more important index
(Schiffman et al., 1981). For this data set, an ordinal measurement
level was chosen as interval level measurement could
not be assumed. The MDS analyses indicate that the stimulus
space is better represented bytwo dimensions (RSQ  0.744;
stress (Kruskaľs stress formula 1)  0.2548) than by three
(RSQ  0.702; stress  0.175). Although the fit improves for
four and five dimensions, using the rule that the square root of
the number of stimuli approximates the number of dimensions
which may reliably be detected by MDS procedures
(Schiffman et al., 1981), two dimensions (
ffiffi
7
p
 2:65)
provide the best solution for this data set. The first dimension
was more important (0.584) than the second one (0.161). The
location of the seven emotion faces within the two-dimensional
space is shown in Fig. 2.
4. Discussion
Overall, the results supported the hypothesis of a distinct
and identifiable pain face. The facial expression of pain
presented on a photo was correctly identified by most participants
from a range of negative emotion expressions.
However, pain expression had the lowest correct identification
rate (unbiased hit rate 58.8%, compared to the best,
91.9% for sadness) and the lowest rating of confidence in
judgements (median 7/10 vs. 9/10 for sadness). Two findings
suggest that participants were able to make a reasonable
estimate of the accuracy of their judgements. First, there
was a substantial correlation between participants' confidence
in their ratings and their accuracy (unbiased hit
rates), across all expressions. Second, confidence ratings
were lower for those who identified the expression incorrectly
than for those who identified it correctly, with the
J. Kappesser, A.C. de C. Williams / Pain 99 (2002) 197­206202
Table 5
Mean, standard deviation, median and range for the confidence ratings
Photo Confidence
Mean (SD) Median Range
Sadness 8.53 (1.67) 9 2­10
Embarrassment 7.77 (2.00) 8 3­10
Surprise 7.75 (1.82) 8 2­10
Anger 7.67 (1.82) 8 3­10
Disgust 7.65 (1.96) 8 3­10
Fear 7.27 (1.96) 8 2­10
Pain 6.65 (2.35) 7 2­10
J. Kappesser, A.C. de C. Williams / Pain 99 (2002) 197­206 203
Fig. 1. Frequencies of confidence ratings according to correct or incorrect photo identification.
largest differences for fear and pain, for which the hit rate was
lowest.
Misidentification of pain for other emotions (disgust,
embarrassment, and fear but not sadness, anger or surprise)
differed from the misidentification of other emotions for
pain (anger, fear and disgust). Except for the (non-significant)
misidentification of pain as disgust, by 18% of participants,
the errors involved small numbers whose
interpretation is injudicious. This is strikingly similar to
the findings by other research groups. Keltner and Buswell
(1996) found that 19.5% of subjects misidentified pain as
disgust (with 52% identifying it correctly and no other
major misidentifications). Hale and Hadjistavropoulos
(1997) found a high incidence of disgust expressions
when videotaping patients during venepuncture and analysing
their faces using the FACS. One reason for the misidentification
could be that disgust and pain share some facial
action units (9 and 10, nose wrinkler and upper lip raiser),
but while these are the only prototypical facial action units
for disgust (Ekman and Friesen, 1978), they occur in combination
with several others in pain (Prkachin, 1992). It is
possible that for those individuals who misidentified pain
as disgust, those common action units were given undue
weight. Misidentification of pain for disgust may also be
attributable to particular characteristics of the pain photo,
which was the same in this study as in Keltner and
Buswelľs study. Since the facial expression of pain is
often blended with other emotions, among these emotions
with disgust, the misidentification may also be explicable by
the frequent co-occurrence of pain and disgust in the face.
Moreover, in terms of the information both facial expressions
convey, both appear to share the motivation to expel
bodily experiences which are not desirable.
Given sufficiently accurate identification of pain expression
and confidence in the judgement, the last part of the
study addressed a separate issue, the perception of negative
emotion expressions in relation to pain, using MDS. Interpretation
of results from only one model can only be
cautious in the absence of replication using a variety of
models. The major axis distinguished embarrassment from
all other expressions, including pain. The obvious difference
between the embarrassed face and all other negative
emotion faces presented is that the embarrassed face is characterised
by AU 12, lip corner puller (Keltner, 1995), producing
a smile of sorts. Interestingly, on viewing the
embarrassed face, several participants remarked that "someone
who smiles cannot be in pain". The second axis distinguished
anger, sadness and pain near one extreme, from
surprise, fear and disgust near the other; embarrassment
was intermediate. The groupings on this dimension cannot
be explained by particular combinations of facial action
units (see Table 1). However, one of the anonymous referees
suggested an alternative basis for the clusters, that of
physical characteristics of the expressions: fear, surprise and
disgust share vertical changes in the face, while pain, anger
and sadness involve more centrifugal changes. This seems
entirely plausible, and no data can be presented here to
distinguish between the two explanations. Speculatively,
one might suggest some meeting ground between the two
explanations in terms of the evolution of multiple facial
expressions from their morphological origins, which
perhaps could be characterised in terms of vertical change
as the origin of one set of related expressions and centrifugal
change as the origin of another set.
An alternative framework for interpreting the second axis
may be the contextual understanding of expression according
to Fridlunďs (1994, 1997) Behavioural Ecology View
of faces: facial expressions as social tools which are `reaď
according to the context of the interaction in which they
occur. For instance, a real smile signals readiness to affiliate,
a false smile signals appeasement. Given that the task
involved imagining that the photo was of a patient presenting
in the ER, the pain face conveys suffering and an attempt
to enlist help, as does the sad face, as indicated by participants'
comments such as "Oh! Poor him!". This explanation
cannot apply to the anger face whose proximity to sadness
and pain remains puzzling. A partial explanation, to do with
the nature of the stimulus, was the comment of a number of
participants that the angry (male) face could represent an
attempt not to show pain, but with difficulty, resulting in a
rigid and staring expression. In contrast, the faces which
express fear, surprise, disgust and embarrassment appear
to represent a reaction to an external stimulus or situation.
Why did MDS place pain and disgust at opposite ends of the
second axis, while the judgement study demonstrated confusion
of the pain face with the expression of disgust by a
substantial minority of participants? The nature of the
tasks ­ judging similarity of pairs of faces to pain versus
naming the specific emotion represented ­ is perhaps more
different than is at first apparent. In the judgement study,
confusion of pain for disgust is assumed to be due to similarity
of the two expressions (as described above in terms of
facial action units), but this is not certain, whereas for MDS,
participants were explicitly asked to judge similarity. Only
finer grained methods, such as asking subjects to explain
J. Kappesser, A.C. de C. Williams / Pain 99 (2002) 197­206204
Fig. 2. Two-dimensional plot of similarities and differences with respect to
pain of the seven facial expressions.
their identification of emotion from facial expression, could
distinguish between possible explanations.
The main methodological problem affecting interpretation
of results is the confounding of characteristics of the
expression with characteristics of the particular photo and
model. These cannot be distinguished within a judgement
study, which uses only one model and set of photographs;
however, this consideration was overruled by the need for
comparability of the stimulus across participants, who were
available only for a short time. It may therefore be the case
that when participants viewed the stimulus photo, they were
comparing the expression shown with an imagined
`perfectly posed expression', as much as or rather than
with an emotion or pain expression category (Etcoff and
Magee, 1992; Young et al., 1997).
A second consideration is that emotional and pain expression
as encountered in clinical and other settings is much
more likely to occur as blends of expressions (Fridlund and
Duchaine, 1996; Hale and Hadjistavropoulos, 1997;
LeResche and Dworkin, 1988; Prkachin, 1997) than as
sequences of prototypes; further, the setting itself provides
an aid to identification and understanding. Both of these are
problems for the interpretation of studies using posed photographs.
In this study, participants were instructed to imagine
the photograph as a patient presenting in the ER, and their
comments suggested that most were able to do so. In relation
to blends of expression, although some researchers
argue that posed expression has no relevance to the interpretation
of spontaneous expression (Motley and Camden,
1988; Russell, 1994), there is some evidence that processing
of expression uses categories. Young et al. (1997) found
categorical recognition of facial expression in photos
which used computer morphing to produce intermediate
blends between posed prototypes.
The simple hit rates in this study were comparable with
those of Keltner and Buswell (1996) and Haidt and Keltner
(1999),whousedthesamephotosbutdifferentjudges(undergraduate
students rather than clinical professionals). Use of
the unbiased hit rate represented a methodological improvement
on the simple hit rate for estimating accuracy of judgements,
but the present sample was not of sufficient size, nor
was the error rate high enough, to investigate the possible
influences of age, sex and experience on judgement. Further
understanding of identification and misidentification of pain
expressionin thecontextof other emotion wouldbenefit from
replicationsusingblendsofexpression,orvideotapeofactual
patients, as stimuli forjudgementsbygroups ofhealth professionals,
patients themselves, and others.
This work has several clinical implications, the more so
because results were generated from a sample of health
professionals. While both students (Prkachin et al., 1994)
and medical professionals (Choiniere et al., 1990; Teske et
al., 1983; Zalon, 1993) tend to underestimate pain as rated
by the patient, there appears to be some influence on underestimation
of exposure to pain expression, although not
straightforward. In a study by Prkachin et al., (2001),
three groups of people who differed in their exposure to
pain estimated pain on videotapes of patients undergoing
a painful medical procedure. By comparison with patients'
pain ratings, observers with experience of pain in their
families attributed greater pain to the patients than did
observers with no experience of pain in their families.
However, health care professionals (physiotherapists and
occupational therapists) attributed least pain to the patients.
It may be that sensitivity to pain expression is enhanced by
repeated exposure to a family member in pain, where judgements
can be confirmed or disconfirmed, but not by multiple
exposure to strangers and little feedback on the quality of
judgements made. An alternative interpretation is that taken
in light of evidence of adequate identification of pain by
health professionals in the present study, this finding
suggests a cognitive discounting bias among health professionals
when faced with a patienťs pain. It is unlikely,
therefore, that training health professionals better to identify
pain (Galin and Thorn, 1993; Solomon et al., 1997) will
significantly improve the match of their pain estimate to
those of the patients. Underestimation bias can be produced
in experimental situations by alerting judges to possible
deception (Poole and Craig, 1992), and in clinical settings
by certain information about the presenting patient, such as
injury type (Todd et al., 1994); it may be better understood by
reference to biases in detection of social cheating (Cosmides,
1989) and lying (Ekman and O'Sullivan, 1991). Systematic
underestimation of pain by health professionals is of particular
concern because of the widespread undertreatment of
pain recorded in many settings (Donovan et al., 1987; Lavies
et al., 1992; Melzack et al., 1987; Owen et al., 1990).
Acknowledgements
This work was completed while the first author was
supported by a Postgraduate Scholarship from the German
Academic Exchange Service, Bonn, Germany. The authors
wish to express their thanks to Dacher Keltner for providing
the photos, to Stephen Morley and Matthew Hankins for
guidance on the use and interpretation of multidimensional
scaling, to Kate Leafhead and Suzanne Skevington for
discussion of ideas, and to the anonymous referees for most
helpful comments. They also thank the staff of Accident and
Emergency (ER) Departments who volunteered to participate
and senior staff for their support: Dr David Williams,
Dr Jane Terris, Andie Howard, Jean Haire, and Chris Ellis.
References
von Baeyer CL, Johnson ME, McMillan MJ. Consequences of nonverbal
expression of pain: patient distress and observer concern. Soc Sci Med
1984;19:1319­1324.
Buck R. Social and emotional functions in facial expression and communication:
the readout hypothesis. Biol Psychol 1994;38:95­115.
Boucher JD. Facial displays of fear, sadness and pain. Percept Motor Skills
1969;28:239­242.
J. Kappesser, A.C. de C. Williams / Pain 99 (2002) 197­206 205
Carroll JM, Russell JA. Do facial expressions signal specific emotions?
Judging emotion from the face in context. J Pers Soc Psychol
1996;70:205­218.
Choiniere M, Melzack R, Girard N, Rondeau J, Paquin MJ. Comparisons
between patients' and nurses' assessment of pain and medication efficacy
in severe burn injuries. Pain 1990;40:143­152.
Cosmides L. The logic of social exchange. Cognition 1989;31:187­276.
Craig KD. Ontogenetic and cultural influences on the expression of pain in
man. In: Kosterlitz HW, Terenius LY, editors. Pain and society, Weinheim:
Verlag Chemie GmbH, 1980. pp. 37­52.
Craig KD. The facial expression of pain. Better than a thousand words? Am
Pain Soc J 1992;1:153­162.
Craig KD, Patrick CJ. Facial expression during induced pain. J Pers Soc
Psychol 1985;48:1080­1091.
Craig KD, Prkachin KM. Nonverbal measures of pain. In: Melzack R,
editor. Pain measurement and assessment, New York, NY: Raven
Press, 1983. pp. 173­179.
Craig KD, Hyde SA, Patrick CJ. Genuine, suppressed and faked facial
behavior during exacerbation of chronic low back pain. Pain
1991;46:161­171.
Craig KD, Prkachin KM, Grunau RVE. The facial expression of pain. In:
Turk DC, Melzack R, editors. Handbook of pain assessment, New York,
NY: Guildford Press, 2001. pp. 153­169.
Donovan M, Dillon P, McGuire L. Incidence and characteristics of pain in a
sample of medical­surgical inpatients. Pain 1987;30:69­78.
Ekman P. An argument for basic emotions. Cognition Emotion
1992;6:169­200.
Ekman P, Friesen WV. Facial action coding system, . Palo Alto, CA:
Consulting Psychologists Press, 1978.
Ekman P, Friesen WV. A new pan cultural expression of emotion. Motivation
Emotion 1986;10:159­168.
Ekman P, O'Sullivan M. Who can catch a liar? Am Psychol 1991;46:913­
920.
Etcoff NL, Magee JJ. Categorical perception of facial expression. Cognition
1992;44:227­240.
Fordyce WE. Behavioural methods for chronic pain and illness. St. Louis,
MO: Mosby, 1976.
Fridlund AJ. Human facial expression: an evolutionary view. San Diego,
CA: Academic Press, 1994.
Fridlund AJ. The new ethology of human facial expressions. In: Russell JA,
Fernandez-Dols JM, editors. The psychology of facial expression,
Cambridge, MA: Cambridge University Press, 1997. pp. 103­129.
Fridlund AJ, Duchaine B. `Facial expressions of emotion' and the delusion
of the hermetic self. In: Harre R, Parrott WG, editors. The emotions:
social, cultural and biological dimensions, London: Sage, 1996. pp.
259­284.
Galin KE, Thorn BE. Unmasking pain: detection of deception in facial
expressions. J Soc Clin Psychol 1993;12:182­197.
Hadjistavropoulos HD, Craig KD. Acute and chronic low back pain: cognitive,
affective, and behavioral dimensions. J Consult Clin Psychol
1994;62:341­349.
Haidt J, Keltner D. Culture and facial expression: open-ended methods find
more expressions and a gradient of recognition. Cognition Emotion
1999;13:225­266.
Hale CJ, Hadjistavropoulos T. Emotional components of pain. Pain Res
Manage 1997;2:217­225.
International Association for the Study of Pain. Pain terms: a list with
definitions and notes on usage. Pain 1979;6:249­252.
Keltner D. Signs of appeasement: evidence for the distinct displays of
embarrassment, amusement, and shame. J Pers Soc Psychol
1995;68:441­454.
Keltner D, Buswell BN. Evidence for the distinctness of embarrassment,
shame, and guilt: a study of recalled antecedents and facial expressions
of emotion. Cognition Emotion 1996;10:155­171.
Lavies N, Hart L, Rounsefell B, Runciman W. Identification of patient,
medical and nursing staff attitudes to postoperative opioid analgesia:
stage 1 of a longitudinal study of postoperative analgesia. Pain
1992;48:313­319.
LeResche L. Facial expression in pain: a study of candid photographs. J
Nonverbal Behav 1982;7:46­56.
LeResche L, Dworkin SF. Facial expression accompanying pain. Soc Sci
Med 1984;19:1325­1330.
LeResche L, Dworkin SF. Facial expressions of pain and emotions in
chronic TMD patients. Pain 1988;35:71­78.
LeResche L, Dworkin SF, Wilson L, Ehrlich KJ. Effect of temporomandibular
disorder pain duration on facial expressions and verbal report of
pain. Pain 1992;51:289­295.
Melzack R, Abbott FV, Zackon W, Mulder DS, Davis MWL. Pain on a
surgical ward: a survey of the duration and intensity of pain and the
effectiveness of medication. Pain 1987;29:67­72.
Motley MT, Camden CT. Facial expression of emotion: a comparison of
posed versus spontaneous expressions in an interpersonal setting. West
J Speech Commun 1988;52:1­22.
Owen H, McMillan V, Rogowski D. Postoperative pain therapy: a survey of
patients' expectations and their experiences. Pain 1990;41:303­307.
Patrick CJ, Craig KD, Prkachin KM. Observer judgements of acute pain:
facial action determinants. J Pers Soc Psychol 1986;50:1291­1298.
Poole GD, Craig KD. Judgements of genuine, suppressed, and faked facial
expressions of pain. J Pers Soc Psychol 1992;63:797­805.
Prkachin KM. The consistency of facial expressions of pain. A comparison
across modalities. Pain 1992;51:297­306.
Prkachin KM, Berzins S, Mercer SR. Encoding and decoding of pain
expressions: a judgement study. Pain 1994;58:253­259.
Prkachin KM, Solomon P, Hwang T, Mercer SR. Does experience affect
judgements of pain behaviour? Evidence from relatives of pain patients
and health-care providers. Pain Res Manage 2001;6:105­112.
Prkachin KM, Currie NA, Craig KD. Judging nonverbal expressions of
pain. Can J Behav Sci 1983;15:409­421.
Prkachin KM, Mercer SR. Pain expression in patients with shoulder pathology:
validity, properties and relationship to sickness impact. Pain
1989;39:257­265.
Russell JA. Is there a universal recognition of emotion from facial expression?
A review of the cross-cultural studies. Psychol Bull
1994;115:102­141.
Schiffman SS, Reynolds ML, Young FW. Introduction to multidimensional
scaling. New York, NY: Academic Press, 1981.
Solomon PE, Prkachin KM, Farewell V. Enhancing sensitivity to facial
expression of pain. Pain 1997;71:279­284.
Teske K, Daut RL, Cleeland CS. Relationships between nurses' observations
and patients' self-reports of pain. Pain 1983;16:289­296.
Todd KH, Lee T, Hoffman JR. The effect of ethnicity on physician estimates
of pain severity in patients with isolated extremity trauma. J Am
Med Assoc 1994;271:925­928.
Wagner HL. On measuring performance in category judgement studies of
nonverbal behavior. J Nonverbal Behav 1993;17:3­28.
Wagner HL. Methods for the study of facial behavior. In: Russell JA,
Fernandez-Dols JM, editors. The psychology of facial expression,
Cambridge, MA: Cambridge University Press, 1997. pp. 31­54.
Young AW, Rowland D, Calder AJ, Etcoff NL, Seth A, Perrett DI. Facial
expression megamix: tests of dimensional and category accounts of
emotion recognition. Cognition 1997;63:271­313.
Zalon ML. Nurses' assessment of postoperative patients' pain. Pain
1993;54:329­334.
J. Kappesser, A.C. de C. Williams / Pain 99 (2002) 197­206206