Observer judgments of acute pain: Facial action determinants

Journal of Personalia and Social Ps\cholog> I486. Vol. 50, No. 6.' 1291 -1298

Copyright 1986 b> (he American Ps>chologicaJ Association. Inc. 0022-.)5M/86/$0075

Observer Judgments of Acute Pain: Facial Action Determinants Christopher J. Patrick and Kenneth D. Craig University of British Columbia, Vancouver, Canada

Kenneth M. Prkachin

This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

University of Waterloo, Waterloo, Ontario, Canada We provided a microanalytic description of facial reactions to a series of painful and nonpainful electric shocks and examined the impact of these as discrete facial cues for observer judgments of acute pain. Thirty female volunteers were videotaped and reported their discomfort in response to electric shocks after earlier exposure to one of three social influence conditions: a tolerant model, an intolerant model, or neutral peer presence. We coded the videotapes for facial activity using the Facial Action CodingSystem (Ekman & Friesen, 1978b). and peer judges rated them for painful discomfort. Subjects exposed to a tolerant model reported no more discomfort than did subjects exposed to an intolerant model, despite receiving more intense levels of shock, but were judged by observers to be in more pain. Analyses of facial activity yielded consistent findings: Tolerant-model subjects, though reporting discomfort equivalent to that reported in other groups, displayed more pain-related facial acti\ ity (brow lowering, narrowing of the e\e aperture from below, raising the upper lip, and blinking). There was a substantial direct relation between observer judgments of distress and discrete, painrelated facial actions (mean multiple R = .74 for the \arious shock levels rated). These data indicate that nonverbal expression yields information about the response to noxious stimulation that is nonredundant with self-report.

Pain can be conceptualized as a construct, verbal report being only one feature of the nomological network of pain experience. Information from other sources may provide important insights into a sufferer's subjective experience, particularly when it is discrepant with self-report. Nonverbal expression offers a promising adjunct to self-report measures of pain (Craig & Prkachin, 1983). Besides being less amenable to conscious distortion (Ekman & Friesen, 1969a, 1974), nonverbal behavior may contribute more to clinical judgments of pain than patient report (Johnson, 1977). Of particular interest is facial expressive behavior, regarded by many to be the primary source of nonverbal affective communication (Ekman & Friesen, 1969b; Rinn. 1984; Tomkins, 1962, 1963). Distinctive movements of the facial musculature have been found to characterize fundamental affective states (Ekman, 1972). Furthermore, observers can use facial cues to make accurate judgments of emotional states (Ekman & Friesen, 1969b; Ekman, Friesen, & Ellsworth, 1983). Recent evidence also indicates that specific facial movements are associated with pain (Craig & Patrick, 1985). The development of a fine-grained assessment technique, the Facial Action Coding System (FACS: Ekman & Friesen, 1978a, 1978b), has made it possible to study the facial expressive correlates of subjective experience. As an objective, anatomically

based system. FACS permits a full description of the basic units of facial movement associated with private experience, including pain. Craig and Patrick (1985) used the system to identify facial activity associated with exposure to a prolonged noxious stimulus, the cold-pressor test. The most involved expressions occurred immediately after the onset of cold-pressor stimulation. Facial activity declined thereafter, despite increases in reported distress, which suggests that facial actions may be most revealing of the immediate impact of discrete noxious events. Another implication was that evidence for a unitary, prototypic pain expression might best be sought in the context of acute noxious stimulation. In support of these conclusions, Prkachin, Currie. and Craig (1983) found that untrained observers providing global judgments were able lo discriminate reactions to painful and nonpainful shocks. Furthermore, biasing instructions led observers to attribute more discomfort to shocked subjects when they were believed to be hypersensitive to pain. Our study was designed to enable us to identify specific facial actions associated with shock-induced acute pain, and we examined the relation of facial actions to untrained observers' pain judgments. Of interest was whether a series of suprapain threshold shocks would be differentially encoded in facial expressive behavior, and whether the frequency of specific facial actions would vary with observer judgments of pain. Of further interest was the relative impact on self-report and facial pain indices of exposure of shocked subjects to a potent source of influence, social modeling (Craig, 1978). Lastly, we reexamined the effect of instructional sets on observer pain judgments.

Preparation of this article was supported by a grant from the Social Sciences and Humanities Research Council of Canada to Kenneth D. Craig and Kenneth M. Prkachin. We gratefully acknowledge the assistance of Brenda Gerhard, Douglas Lee, and Cara Zaskow in the conduct of this study. Correspondence concerning this article should be addressed to Kenneth D. Craig, Department of Psychology, University of British Columbia, Vancouver, British Columbia, Canada V6T 1Y7.

Method The study was conducted in two phases. In the first, three groups of subjects received an ascending series of electric shocks while a confederate

1291

1292

C. PATRICK. K. CRAIG, AND K. PRKACHIN

coparticipant modeled either tolerance or intolerance to the shocks, or remained inactive. Thereafter, we delivered a random series of shocks, sampling varying degrees of discomfort identified during the previous series. Subjects were videotaped throughout. In the second phase, untrained observers w itnessed the videotapes and rated the level of discomfort that they believed the subjects were experiencing. One group of observers was led to believe that the subjects were hypersensitive to pain, and another that the subjects were hvposensitive. A third group received neither instruction.

This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Subjects and Models Thirty female University of British Columbia undergraduates (ages ranged between 17 and 28) were assigned randomly to three experimental groups after agreeing to participate for a small fee. Female subjects were used because they tend to be more overtly expressive (cf. Buck. Miller, & Caul. 1974; Schwartz, Brown, & Ahern. 1980). An equal number of subjects appeared in each of the tolerant-model, intolerant-model, and control conditions. Seven additional subjects (5 tolerant-model and 2 control subjects) exercised their right to withdraw from the study before or during the random shock series. The models were 3 female undergraduates treated as naive subjects but trained to assume roles appropriate to the modeling conditions. Models were assigned randomly to subjects.

Apparatus and Materials Electric currents of .05-s duration were controlled by a Hunter decade interval timer and delivered to the volar surface of the left forearm through concentric annular electrodes (Tursky, Watson, & O'Connell, 1965) from a controlled-current stimulator, in accordance with procedures described by Craig and Prkachin (1978). Subjects were videotaped through a oneway mirror, as described by Craig and Patrick (1985). The subject and model were seated side by side, facing the one-way mirror and separated by a partition. A rectangular 10 X 45 cm response panel was positioned on a wooden frame in front of and between them. On the panel was a series of 14 push-button/jewel-light pairs labeled with affective descriptors in an ascending order, from undeieciable at the extreme left to ICTT intolerable at the extreme right. The latter 12 descriptors had been ratio-scaled bv Gracely, McGrath, and Dubner (1979). The first two descriptors, labeled undeieciable and detectable, but not unpleasant, were included in order to provide information about prepain sensations. The rating panel permitted reports of discomfort during the ascending and random series. One other push-button/jewel light, labeled painful, was set apart from the others on the panel. Subjects used this button during the ascending series to indicate when they first perceived a shock to be painful (i.e., pain threshold). The experimenter recorded ratings and delivered shocks from an adjacent room.

Procedure Phine I . After introductions, the studv was described as an examination of people's reactions to electric shocks. The participants were told that they would receive two series of electric currents, an ascending series and a random series; none of the latter was to exceed the highest level accepted in the ascending series. They were also informed of their right to withdraw at any time and about the videotaping procedure before signing a consent form. The subject and model were then seated side by side facing the oneway mirror in a 3 X 3 m testing chamber. The experimenter attached a shock electrode to each person's arm farthest from the screen, leaving her inner arm free to use the rating panel. Despite identical procedures for both participants, only the subject's electrode was functional. The procedure for the tolerant- and intolerant-model groups was described as follows: The participants would receive an ascending series of

electric shocks beginning at undetectable levels and gradually increasing in intensity. A clicking sound from the timer would signal the delivery of each stimulus. Immediately after each shock, the participants were to rate their discomfort on the 14-point scale; the subject would rate first. Descriptors higher on the panel were to be used to denote increasing levels of discomfort. When a participant received the first shock of the series that she felt was painful, she was to push the button marked painful before rating her discomfort on the 14-point scale. It was emphasized that the button labeled painful was to be used only once. Thereafter, the subject would continue making ratings on the 14-point scale until she reached a level of shock at which she wished to stop. The participants were told they could stop the ascending series at any time by pushing the highest button on the rating scale, marked very intolerable. In the tolerant-model group, the model made lower ratings of discomfort than the subject during the ascending series. The model generally remained two descriptors behind on the rating scale, used the painful button before selecting a rating that was one higher than that chosen by the subject after the latter hit the painful button, and selected the highest rating on the panel only after the subject had withdrawn from the ascending series. In the intolerant-model group, the model generally remained two descriptors ahead of the subject (meaning the former terminated the series first), and hit the painful button before using the fifth descriptor (unpleasant) on the 14-point rating scale. The details of the tolerant and intolerant modeling roles were comparable with those described b> Craig. Best, and Ward (1975). In the control group, the model remained inactive during the ascending series. After describing the rating procedure, the experimenter explained that the shock equipment was designed to accommodate only one subject at a time. The model was asked to remain seated while the subject participated first. The control group instructions were identical to those for the other groups except that they made no reference to concurrent participation. The experimenter then administered the ascending series. Current intensities increased in 0.5-mA steps from 0.0 to a maximum of 16.0 mA, with a mean interstimulus interval of 20 s (range 15-25 s). The subject was videotaped throughout the ascending series and also for 30 s before delivery of the first shock; the latter provided a neutral expression standard for FACS coding. After the ascending series, the experimenter returned to the experimental chamber to discuss the random series. The subject was to accept her random shocks first and to rate the stimuli as before, except that the painful button would not be used. The random series consisted of 15 shocks, 3 at each of five different intensity levels. The first was two thirds of the intensity of the shock identified as painful (pain threshold) in the previous series. The second and fifth levels matched the subject's pain threshold shock level and the magnitude of her terminal ascending-series shock (tolerance), respectively. Shock levels 3 and 4 were current intensities one third and two thirds between pain threshold and tolerance levels. Three subjects ( I in the intolerant-model and 2 in the tolerant-model groups) did not use the painful button at all during the ascending series. Six other subjects (2 control subjects and 4 in the tolerant-model groups) used the painful button at or so near their terminal ascending-series shock level (i.e., within 0.5 mA) that it was not feasible to select random-series shocks varying between actual threshold and tolerant levels. For these 9 subjects, random-series shock levels were selected differently. Shock Level 5 remained at pain tolerance level, but Shock Levels 2-4 were set at 1.5, 1.0. and 0.5 mA below Level 5, respectively. This was done in order to establish a degree of discriminability among these four shock levels. Shock Level I was two thirds the magnitude of the arbitrarily designated "threshold" level (Level 2), in accordance with the procedure used for the majority of subjects. The 15 shocks in this series were delivered in random order, with the stipulation that no two shocks of identical magnitude were administered consecutively. In order to control for order effects, the relative magnitude

FACIAL EXPRESSION OF ACUTE PAIN

1293

Table I Group Means and Standard Deviations for Current Intensities (mA) Delivered at Each Random-Series Shock Level Random-series shock level

Group

M

SD

M

SD

M

SD

M

SD

M

SD

Tolerant Control Intolerant

6.23 4.38 3.45

2.16 2.21 1.83

9.30 6.60 5.20

3.22 3.37 2.71

10.05 7.58 5.86

3.24 3.50 2.71

10.80 8.55 6.55

3.38 3.70 2.83

11.55 9.50 7.15

3.58 4.14 2.96

This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Note, n = 10 cases per group: current intensities for each case represent the means for the three shocks delivered at each level.

of the first shock of the random series varied from subject to subject in random fashion, but was balanced both within and between groups. After the random series, subjects were fully debriefed as to the purpose of the study. Phase 2. In this phase, 30 female undergraduate observers (aged 1634) were paid for participating in two 2-hr sessions in which they veiwed videotapes of subjects from Phase 1 and made ratings of their discomfort. Groups of 2-6 observers watched the videotapes on a 23-in. black and white television screen and recorded their ratings on a coding form. There were two videotapes, each showing one half (n = 15) of the subjects from Phase 1. Each observer viewed and rated both videotapes twice in sessions separated by I week. Viewing order was counterbalanced across sessions. Each subject was shown in turn receiving a series of 18 trials, the 15 random-series shocks accepted by that subject plus her first ascendingseries trial (0.0 mA) repeated three times. The latter were interspersed randomly among the former. Shock segments (18 per subject) were separated by 3-s periods of blank screen in order to allow observers to record their ratings. A clicking sound was dubbed over each segment in order to designate the exact point at which the shock was delivered. Observers were told that the study concerned the ability of people to "identify and judge the expressive behavior of other people who are experiencing various levels of pain." As in an earlier judgment study (Prkachin et al., 1983), three types of instructions regarding the content of the videotapes were given. One group of 10 observers ("hypersensitive") was told that the subjects on tape were receiving various levels of electric shock after having had their arms abraded to make them hypersensitive to pain. A second group ("analgesic"; n = 10) was told that the subjects had received a pain-relieving drug before accepting the shocks. The third group ("control"; « = 10) received no mention of any preshock manipulation. Observers were assigned randomly to instructional conditions. Observers watched the screen and. after each click signaling a shock, rated how much pain the subject felt. Ratings were based on a 7-point scale denoting increasing levels of discomfort from 0 (subject felt nothing) to 6 (strongpain). To prepare themselves for the rating task, the observers completed three practice trials in which they rated reactions to 18 shocks by each of three subjects; the practice subjects were not part of the 30subject sample subsequently scored.

Facial Action Coding Using FACS (Ekman & Friesen. 1978b). we scored subjects' reactions to the 15 random-series shocks. For each random-series shock, we scored a 3-s segment of videotape that consisted of the 0.5-s period preceding the shock, the 0.5-s shock itself, and the 2-s period following the shock. These were the same stimuli witnessed by observers in Phase 2. In addition, videotape segments capturing subjects' reactions to the first three shocks of the ascending series and to the final shock of that series were scored. The former provided a baseline against which to compare reactions to the more intense random series stimuli.

The two data coders had passed proficiencv tests as FACS scorers (Ekman & Friesen, 1978a). They were blind to the group membership of stimulus subjects and to current intensities that were being delivered. One coder scored all videotape segments, and the other provided reliability scoring on 20t"c> of the data (i.e., four random 1> selected segments per subject).

Results Noxious Stimuli Delivered Current intensities for Shock Levels 1-5 of the random series were compared across the three modeling groups in a 3 X 5 (Modeling Group x Shock Level) analysis of variance (ANOVA) with repeated measures. The group means for each level appear in Table 1. The main effect of modeling group was significant, F(2, 108) = 4.41, p < .05; Newman-Keuls comparisons (a = .05) revealed a significant difference between the means for the tolerant- and intolerant-model groups. The overall shock level effect was also significant, F\4,108)= 112.89, p< .0001; multiple comparisons indicated that all means differed from one another. The Modeling Group X Shock Level interaction was not significant. These results indicated that the current intensities for the three experimental groups increased progressively from Shock Level 1 to 5; the tolerant-model subjects received higher shocks at each level than did the intolerant-model subjects. Because Shock Level 5 for all subjects matched the magnitude of their terminal ascending-series shock, one also can conclude that tolerant-model subjects exhibited greater pain tolerance than did intolerant-model subjects. The same may be said in regard to pain threshold: The current delivered at Shock Level 2 was equivalent to the subject's ascending series pain threshold, except in cases in which the threshold was not reached or was within 0.5 mA of the subject's tolerance level. In this situation, which was the case for 6 tolerant-model subjects and only 1 intolerantmodel subject, Shock Level 2 was an underestimate of pain threshold. Still, Shock Level 2 was higher for subjects exposed to tolerant models, so that one may safely conclude that subjects in this group exhibited a higher threshold for pain.

Self-Reported-Distress Subjects' discomfort ratings at the five random-series shock levels were examined as a function of the modeling manipulation. Ratings were converted to their ratio-scale values (Gracely et al., 1979). The first and second descriptors on the rating scale (un-

This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

1294

C. PATRICK. KL. CRAIG, AND K. PRKACHIN

detectable and detectable but not unpleasant) were not part of the Gracely et al. scale: ratings corresponding to these two descriptors were assigned numerical values of 0 and 1.4, respectively, the latter representing the midpoint between 0 and the scale value of the first Gracely et al. descriptor (slightly unpleasant). The analysis was 3 X 5 repeated-measures A NOVA: modeling group and shock level were the between- and w ithin-group factors, respectively. The dependent measure was the mean rating score for the three shocks delivered at each level. The overall effect of shock level was significant, F(4, 108) = 31.43. p < .005; NewmanKeuls comparisons revealed significant differences among all pairs of means except those for shock levels 2 and 3. The modeling group main effect and Modeling Group X Shock Level interaction effect were nonsignificant. This result indicated that subjects reported greater discomfort as the random-series shocks increased in intensity. Furthermore, tolerant-model subjects reported equivalent degrees of pain despite receiving higher shocks at each level than did intolerant-model subjects.

Observer Ratings A 3 x 3 X 6 (Judge Instruction X Modeling Group x Shock Level) repeated-measures A NOVA was used to evaluate the effect of the experimental manipulations on observer judgments of pain. Ratings of individuals were collapsed within each modeling condition in order to provide one set of observer ratings per group. "Shock level" refers to the six different levels of shock per subject witnessed by judges, collapsed over the three presentations of each shock level and the two separate videotape viewings. The first ascending-series shock was designated Shock Level I , and the five random-series levels were designated 2-6 in order of increasing magnitude. The main effect of shock level \vas significant, F(5. 135) = 762.00, p < .0001: Newman-Keuls comparisons revealed significant differences among all pairs of means except those for Shock Levels 3 and 4 (i.e., random series Shock Levels 2 and 3). This indicated that observers generally judged subjects receiving higher current intensities to be in more pain, irrespective of modeling group assignments and instructional conditions. The judge instruction main effect was not significant, but the Judge Instruction X Shock Level interaction was. F(IO. 135) = 3.18, p < .005. Simple main effects were tested at an alpha level of .05/6 (.0083) in order to maintain the same per family error rate as that used for the overall F test (Kirk, 1968). Using this criterion, we found the judge instruction effect to be significant at only the two highest shock levels. Multiple comparisons in which we used the Newman-Keuls procedure (a = .05) revealed a significant difference between means at Shock Level 6, at which observers who received the analgesic instructions gave higher pain ratings than did those who heard the hypersensitive instructions. The main effect of modeling group was significant, F(2, 54) = 308.90, p < .0001: Newman-Keuls tests indicated that all group means differed from one another. Judges tended to attribute most pain to tolerant-model subjects, less pain to controls, and the least pain to intolerant-model subjects (Af = 3.16. 2.60, and 2.34, respectively). The Modeling Group X Shock level interaction effect was also significant, F( 10,270) = 128.06, p < .0001. Tests of simple main effects (a = .0083) revealed significant

modeling group differences at all six levels of shock, and Newman-Keuls comparisons indicated that pain ratings for tolerantmodel subjects exceeded those for the other groups at all but the first shock level. Also, in regard to the upper five shock levels, control subjects were judged to be in more pain than were the intolerant-model subjects except at Level 4. at which pain ratings for the latter barely exceeded those for controls. These results reflect a discrepancy between observers'judgments of pain and subjects' self-reported discomfort. Despite receiving more intense shocks than intolerant-model subjects at the five shock levels that constituted the random series, tolerant-model subjects did not report greater discomfort, although judges perceived them to be in greater pain.

Subject Facial Expressions Intercoder reliability for FACS frequency scoring of facial action units (AUs) was calculated according to the criteria used by Craig and Patrick (1985). The ratio of the total number of agreements to the total number of AUs scored was .74. Intercoder agreement for the subset of AUs found to reflect pain (to be described shortly) was .84. These figures closely resemble those reported by Craig and Patrick. Data reduction techniques were used at the outset in order to isolate facial actions associated with shock-induced pain. AUs that showed a mean frequency of less than three occurrences for the sample as a whole (/V = 30) over the six shock levels coded (baseline plus the five random-series levels) were not treated further because of their rarity (i.e., incidence was less than 18/180). This left 13 AUs. In repeated-measures Hotelling's T2 analyses, we compared facial activity during baseline with the mean incidence of AUs provoked by the strongest shocks delivered (i.e., the last shock of the ascending series and the three delivered at Level 5 of the random series). In an initial analysis we examined whether similar AUs were characteristic of both cold-pressor and shock-induced pain. The dependent measures for the analysis were the 7 AUs out of the 13 just mentioned that had been found to be associated with cold-pressor pain: AUs 6, 7, 10. 12, 25, 26, and 45 (Craig & Patrick, 1985). The overall test was significant, F(7, 173) = 8.64, p < .0001. Multiple comparisons at the .05 alpha level in which we used the Bonferroni (BON) procedure to control for experimentwise error (Ramsey, 1980) revealed that three of the seven AUs occurred significantly more often during painful stimulation than during baseline: AUs 6, 10, and 45 (see Table 2). A fourth variable, AU 7, merely approached a conventional level of significance (p < .07). We repeated the same analysis, using all 13 AUs to determine whether any additional facial actions were prominent during shock-induced pain. Again, the overall test was significant, F( 13. 167) = 4.84, p < .0001; BON comparisons revealed significant effects for four individual AUs: AUs 6, 10, and 45, as identified in the first analysis, plus AU 4 (see Table 2). In a further analysis we examined modeling group differences in subjects' facial activity at the five random-series shock levels. We conducted a 3 X 5 (ModelingGroup X Shock Level) repeatedmeasures ANOVA, using the sum of the frequency scores for the four AUs strongly associated with pain (AUs 4, 6, 10, and 45, henceforth referred to as "pain-related AUs") as the dependent

FACIAL EXPRESSION OF ACUTE PAIN

This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Table 2 AUs Discriminating Between Electric Shock and Cold Pressor Pain

All

Characteristic of cold pressor pain

Characteristic of electric shock pain

4 (brow lower) 6/7 (cheek raise/lids tight) 10 (upper lip raise) 12 (lip corner pull) 25 (lips part) 26/27 (jaw drop/lips parted) 43/45 (eyes closed/blink)

a

Yes Yes Yes Yes Yes Yes

Yes All 6 alone" Yes No No No AU 45 alone

Note. AU = action unit. AUs 6/7. 26/27. and 43/45 were treated as composite AUs in the cold pressor study by Craig and Patrick (1985). •Marginally significant (p < .10). b AU 7 was marginally significant (p < .07).

measure. The main effect of modeling group was significant, F(2, 27) = 6.68, p < .005; Newman-Keuls comparisons (« = .05) indicated that pain-related AUs were exhibited more frequently by tolerant-model subjects than by control or intolerantmodel subjects (see Table 3). Although the mean occurrence for intolerant-model subjects was lower than for controls, the difference was not statistically significant. The main effect of shock level was also significant, F(4, 108) = 4.29, p < .005: pain AUs were generally more frequent at higher shock levels (see Table 4). Newman-Keuls tests revealed statistically significant differences between the mean for Shock Level 1 and the means for Levels 2, 4, and 5. The Modeling Group X Shock Level interaction was not significant. These results provided indirect evidence that judges' ratings were based on subject facial activity that was found independently to be associated with pain.

Observer Judgments and Facial Activity To examine in a direct way the relation between observer judgments of pain and subject facial expression during the random shock series, we performed stepwise multiple regression analyses. The predictors for each analysis were frequency scores for the four pain-related AUs identified through the FACS system; the criterion scores were mean observer ratings of subjects (N = 30) at each of the five random-series shock levels. A regression equation containing all four predictor variables was significant at each shock level (p < .005); the mean multiple R achieved a

1295

magnitude of .74 (range = .67-.79). AUs 4 and 45 were generally weighted most heavily in the equation; AUs 6 and 10 made little independent contribution, owing to their sizable correlations with AU 4 (see the following paragraphs). The results of this analysis indicated that a substantial proportion of the variance in judges' ratings (55%, on average) was attributable to changes in specific components of subject facial expression. In further analyses we examined whether facial activity would be predictive of self-reported discomfort associated with shocks. The results of an earlier study (Craig & Patrick, 1985) suggested that such a relation, though negligible during more prolonged noxious stimulation, might be evident with phasic acute pain. In accordance with this hypothesis, group means for both selfreport and facial indices of discomfort were higher at more intense shock levels. In multiple regression analyses, in which we used frequency scores for the four pain-related AUs to "predict" selfreported discomfort at each random series shock level, we examined the same relation on a subject-by-subject basis. The mean magnitude of the multiple R was .43 (range = .28-.S5), which was consistent with the magnitude of the relation reported by Craig and Patrick for subjects' immediate reactions to cold-pressor stimulation. We calculated Pearson correlations to examine interrelations among the four pain-related AUs at each shock level of the random series. Earlier research (Patrick, 1983) had suggested that a unitary pain expression comprising two or more AUs in combination would be more likely to appear as a reaction to acute (as opposed to more prolonged) noxious stimulation. In accordance with this prediction, intercorrelations among AUs 4, 6, and 10 at those shock levels that could be considered painful (Levels 2-5) were all substantial and significant (see Table 5 for mean correlations). In Figure 1 we provide a drawing of a single subject's neutral expression and her reaction to painful shock, involving AUs 4, 6. and 10 in combination. Relations between AU 45 and the other pain-related AUs were generally smaller and nonsignificant; the only significant correlations involving AU 45 were with AU 10 at Shock Levels 4 and 5 (re = .31 and .32, respectively, p < .05). At Shock Level 1, at which subjects received sub-pain-threshold currents, correlations between AUs 4 and 6, 4 and 10, and 6 and 10 were .10 (p > .05), .48 (p < .005), and .45 (p < .01), respectively, correlations involving AU 45 were all small and nonsignificant. Taken together, these findings suggest that a specific cluster of facial actions may emerge in response to noxious stimulation, particularly at levels that are characterized as painful. However, these correlations indicate not necessarily a high degree of co-occurrence of these AUs, but

Table 3 Mean Pain-Related Ai' Frequency in Experimental Groups

AU4 Group

Tolerant Control Intolerant

M

0.187

0.113 0.093

All 6

SD 0.391 0.318 0.292

M

0.167 0.020 0.040

AU 10

SD 0.374 0.140 0.197

M

0.247 0.040 0.073

AU45

SD 0.448 0.197 0.309

M

1.873 1.453 0.853

Sum of pain AUs SD

1.101 0.931 0.922

\f

SD

2.473 1.627 1.060

1.549 1.046 1.142

Noie. All = action unit, n = 150 observations per cell (10 Subjects/Group x 5 Shock Levels/Subject x 3 Shocks/Level). Modeling Group X Shock Level interaction was not significant.

1296

C. PATRICK, K. CRAIG, AND K. PRKACHIN

Table 4 Mean Pain-Related AL' Frequency Over Random-Series Shock Levels AU6

AIM

Shock level

M

SD

M

SD

M

SD

1

0.078 0. 1 33 0.089 0.156 0.200

0.269 0.342 0.286 0.364 0.402

0.022 0.100 0.044 0.078 0.133

0.148 0.302 0.207 0.269 0.342

0.067 0.144 0.089 0.133 0.167

0.251 0.412 0.323 0.342 0.375

1

3 4

5

Sum of pain AUs

AU45

AU 10

.222 .478

.333 .500 .433

SD

M

SD

1.014 1.062 1.039 1.124 1.112

.389 .856 .556 .867 .933

.109 .395 .282 .470 .592

This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Note. AU = action unit, n = 90 observations per cell (30 Subjects X 3 Shocks/Level).

only a substantial commonality of appearance together within the same 3-s segment. The data, however, did reveal a high degree of actual co-occurrence. AUs 4, 6, and 10 appeared a total of 59, 34, and 54 times during the random series, respectively. AU 4 appeared concurrently with AU 6 or AU 10 or both 34 times during the random series; AU 6 appeared with at least one of the other two pain-related AUs in question 28 times; and AU 10 appeared as part of a pain-related AU combination 33 times (see Table 6). AU 45, on the other hand, participated in the aforementioned combinations on only nine occasions (five of these at the very beginning or very end of the combination in question), despite appearing a total of 627 times during the random series. As postulated by Craig and Patrick (1985), AU 45 may represent a form of reflexive coping behavior that appears with increased frequency during any ordeal, independent of other facial activity. Also noteworthy was the finding that AU 7, which approached a conventional level of significance in preliminary analyses, tended to accompany AU 4 or AU 10 or both when the latter were not paired with AU 6 (the FACS system does not permit AUs 6 and 7 to be scored concurrently). AU 7 appeared concurrently with AU 4 a total of 21 times and with AU 10 a total of 13 times. These findings suggest that a unitary facial expression, consisting of AUs 4 and 10 in combination with AU 6 or 7, may constitute a typical reaction to acute noxious stimulation.

Discussion The findings attest to the importance of assessing the complex construct of pain through multiple referents. In accordance with

Table 5 Mean Correlations Among Pain-Related AUs (Random Shock Levels 2-5) AU

4

6

10

45

4 6 10 45

-

0.56" —

0.52* 0.65*** —

0.14 0.23 0.21

—

Note. AU = action unit. Correlations are based on mean AU frequency scores (averaged over the three shocks/level) for subjects (n = 30) during random series Shock Levels 2-5; reported significance level is for the smallest correlation contributing to a particular mean r. *p nine occasions, despite appearing a total of 627 times during the random series.

scription of posed facial expressions of pain reported by Hjortsjo (1969). It seems probable that the "physically hurt, tormented" expressions adopted by Hjortsjo's actors are more likely to be observed with acute, intense noxious stimulation than with more extended, dull, aching pain (Craig & Patrick, 1985). The fact that Hjortsjo's expressions included additional components not observed in our stud>—depression of the lip corners, clenching of the jaw. wrinkling of the nose, and squinting of the eyes— suggests that Hjortsjo may have been talking about a response to levels of noxious stimulation that for ethical reasons cannot be simulated in a laboratory setting. Clinical investigations of severe acute pain may yield reaction patterns closer to Hjortsjo's predictions, as was suggested by data from LeResche (1982), who coded naturalistic photographs of people in intense pain. Previous investigations (Craig & Patrick, 1985: Prkachin et al., 1983) suggested that a clear relation between self-report and facial expression, and evidence for a unitary facial reaction pattern, would be observed during acutel> noxious stimulation. Both hypotheses were supported in our study. A mean multiple correlation of .43 between pain-related facial behavior and selfreported discomfort was observed across random-series shock levels. Pain-related facial actions occurred frequently in combination, supporting the position that a unitary facial expression would be a characteristic response to acute pain. The clear presence of pain-related facial AUs permitted a delineation of criteria for observer judgments of pain. Direct evidence that observers use specific facial cues to judge pain was provided by a mean multiple R of .74 between observer ratings and pain-related AU frequency. The magnitude of this relation is remarkable when considered in the light of individual differences in expressive pain behavior and the role that idiosyncratic response biases play in observer judgments of pain (Prkachin et al., 1983). In general, the largest proportion of the variance was accounted for by AUs 4 (brow lower) and 45 (blink): AUs 10

This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

1298

C. PATRICK. K. CRAIG. AND K. PRKACHIN

(upper lip raise) and 6 (cheek raise) were less important, but involved. When considered in conjunction with the finding that pain-related facial activity was more frequent at higher shock levels, our results support the positions that facial actions encode characteristics of noxious stimulus events and that observers can make relatively accurate judgments of these stimulus events and related experiences by observing facial expressions. The observed discrepancy between self-report and facial indices of pain deserves further comment. One might be tempted to infer that the social modeling manipulation caused subjects to falsify reports of their experience. However, such an inference would not be necessary. Pain is a complex construct with multiple referents, and there is no reason to expect that different pain indices should always be in agreement. Desynchrony among alternate measures of pain may provide useful information about a patient's overall experience, as has been the case with anxieU (Craske & Craig, 1984). Our results suggest that facial expressive behavior may represent a nonvoluntary reaction to increasing intensities of noxious stimulation that, though not entirely independent, is not perfectly correlated with subjective experience. The relatively moderate degree of the relation (mean R = .43) between self-report and facial activity suggests that the two measures reflected different aspects of a common reaction pattern. In summary, the value of nonverbal expression may be to provide an index of the objective intensity of noxious stimulation's impinging on an individual. Discrepancies between nonverbal cues and self-reported experience should alert clinicians and others who witness people in pain to psychological, situational, and other variables that mediate pain report.

References Buck, R.. Miller, R., & Caul, W. (1974). Sex, personality, and physiological variables in the communication of affect via facial expression. Journal of Personality and Social Psychology, JO, 587-596. Craig, K. D. (1978). Social modeling influences on pain. In R. A. Sternbach (Ed.), The psychology of pain (pp. 73-109). New York: Ra\en. Craig, K. D., Best. H., & Ward, L. M. (1975). Social modeling influences on psychophysical judgments of eletrical stimulation. Journal of Abnormal Psychology, 84. 366-373. Craig, K. D., & Patrick, C. J. (1985). Facial expression during induced pain. Journal of Personality and Social Psychology. 48. 1080-1091. Craig, K. D.. & Prkachin. K. M. (1978). Social modeling influences on sensory decision theory and psychophysiological indexes of pain. Journal of Personality and Social Psychology, 36. 805-815. Craig, K. D., & Prkachin, K. M. (1980). Social influences on public and private components of pain. In I. G. Sarason & C. D. Spielberger (Eds.), Stress and anxiety (Ibl ?) (pp. 57-72). New \brk: Hemisphere. Craig, K. D., & Prkachin, K. M. (1983). Nonverbal measures of pain. In R. Melzack (Ed.), Pain measurement and assessment (pp. 173179). New York: Raven Press. Craske. M. G., & Craig. K. D. (1984). Musical performance anxiety:

The three systems model and self-efficacy theory. Behaviour Research and Therapy. 22, 267-280. Ekman, P. (1972). Universals and cultural differences in facial expressions of emotion. In J. Cole (Ed.), Nebraska Symposium on Motivation, I9?l, (Vol. 19. pp. 207-281). Lincoln: University of Nebraska Press. Ekman, P., & Friesen, W. V. (1969a). Nonverbal leakage and clues to deception. Psychiatry. 32, 88-106. Ekman, P., & Friesen, W. V. (1969b). The repertoire of nonverbal behavior—Categories, origins, usage, and coding. Semioiica, 1, 49-98. Ekman, P.. & Friesen, W. V. (1974). Detecting deception from the body or face. Journal of Personality and Social Psychology, 29, 288-298. Ekman, P., & Friesen, W. V. (1978a). Investigator's guide to the Facial Action Coding System. Palo Alto. CA: Consulting Psychologists Press. Ekman, P.. & Friesen, W. V. (1978b). Manual for the Facial Action Coding System. Palo Alto, C.A: Consulting Psychologists Press. Ekman, P., Friesen, W. V., & Ellsworth, P. (1983). What components of facial behavior are related to observers' judgments of emotion? In P. Ekman (Ed.). Emotion in the human face (2nd ed.. pp. 98-110). Cambridge. England: Cambridge University Press. Gracely. R. H.. McGrath, P.. & Dubner. R. (1979). Narcotic analgesia: Fen tan yl reduces the intensity but not the unpleasantness of painful tooth pulp sensations. Science. 203, 1261-1263. Harris, G.. & Rollman. G. B. (1983). The validity of experimental pain measures. Pain. I7, 369-376. Hjortsjo, C. H. (1969). Man's face and mimic language. Lung, Sweden: Student Litteratur. Johnson. M. (1977). Assessment of clinical pain. In A. K.. Jacox (Ed.), Pain: A sourcebook for nurses and other health professionals (pp. 451 476). Boston: Little, Brown.

Kirk, R. E. (1968). Experimental design: Procedures for the behavioral sciences. Belmont. CA: Brooks/Cole. LeResche. L. (1982). Facial behavior in pain: A study of candid photographs. Journal of Nonverbal Behavior. 7, 46-56. Patrick, C. J. (1983). Social modeling influences on self-report and facial expression indices oj cold-induced pain. Unpublished master's thesis. University of British Columbia, Vancouver, Canada. Prkachin. K. M.. Currie. A. N.. & Craig. K. D. (1983). Judging nonverbal expressions of pain. Canadian Journal of Behavioural Science, 15, 409-421. Ramse>, P. H. (1980). Choosing the most powerful pairwise multiple comparison procedure in multivariate analvsis of variance. Journal of Applied Psychology, 65. 317-326. Rinn, W. E. (1984). The neurops\chology of facial expression: A review of the neurological and psychological mechanisms for producing facial expressions. Psychological Bulletin. 95, 52-77. Schwartz, G. E., Brown, S., & Ahern, G. L. (1980). Facial muscle patterning and subjective experience during affective imagery: Sex differences. Psycliopliysiology, I?. 75-82. Tomkins, S. S. (1962). Affect, imagery, consciousness. Ibl. I The positive affects. New York: Springer. Tomkins, S. S. (1963). Affect, imagery, consciousness, Ibl. 2. The negative affects. New \brk: Springer. Tursky, B., Watson, P. D.. &O'Connell. D. N. (1965). A concentric shock electrode for pain stimulation. Psychoptiysiology, I. 196-198. Received April 23, 1984 Revision received November 26, 1984 •