Results

Deliberate Actions

Laterality of Deliberate Actions

Verbal versus Visual Tables 1 and 2 show the symmetry of subjects by category, right, left, and symmetrical, for each AU scored. Table 1 shows the symmetry of subjects for the verbally and visually requested actions separately and for these two conditions combined. Individual actions requested in both verbal and visual conditions (AUs 1+2, 4, 7, 9, 10, 12, 20) were combined and averaged together because there were no differences in asymmetry between these conditions. Correlations of asymmetry scores for the same AU in the two conditions were all significant. Patterns of correlations between either the verbal or the visual scores and the other scores described below were the same, although not necessarily at the same significance level. Combining scores for the same AU from the verbally and visually requested actions probably improved the sampling of bilateral requested actions to give a better estimate of the asymmetry of these actions. Table 2 shows the breakdown for these combined scores again and for the scores for "better unilateral control", "preferred side" and "rated easier" scores.

Laterality by AU Inspection of Tables 1 and 2 confirms the hypothesis that different muscle actions show different patterns of laterality. For three of the four measures in Table 2 (excepting the ratings), asymmetry of some actions was lateralized, while other actions showed little tendency for lateralization. For two of the asymmetry measures, bilateral asymmetry and better unilateral control, laterality of some AUs was opposite. For the bilateral asymmetry measure, nominally more AUs that were lateralized were stronger on the right than on the left, partially disconfirming the prediction of left lateralization for these actions. The asymmetry scores for requested bilateral movements in Table 1 indicate that AUs 4 (brow lowered) and 12 (smiling) were lateralized stronger on the left as predicted, but AUs 9 (nose wrinkle), 15 (lip corners down), and 20 (lip stretched to side) were lateralized right. The combination of 1+2 in the brow raise was also lateralized stronger on the right, although the separate scores for AUs 1 (inner) and 2 (outer) suggested that AU 2 was primarily responsible for this effect. The duration of eyes closed during the blink was also lateralized such that the right eye stayed closed longer, partially confirming the hypothesis of right laterality for actions related to blinking. Lateralization was not evident in bilateral asymmetry scores for AUs 1, 6, 7, 6+7, 10, 16, and 45 (strength of contraction during the blink).

Better unilateral control scores also showed laterality and, like bilateral asymmetry scores, laterality changed depending upon the action. Table 2 shows that the smile (AU 12) was lateralized with better actions on the left and the lip stretch (AU 20) tended in this direction. The brow raise (AUs 1+2) was lateralized with better actions on the right.

Table 2 shows that rated easier and preferred side scores are different from the other two measures in that no left laterality was observed. Nominally more subjects rated the right side easier than the left for nine of the ten actions, but none of these ratings for individual AUs showed significant laterality although AUs 2 (outer brow raise) and 4 (brow lowered) showed strong tendencies. Likewise, scores for the preferred unilateral movement showed that more subjects preferred moving their right than their left side for all AUs except for AUs 1 and 4 which showed a few more left than right subjects, but only two individual actions were significantly lateralized. Subjects preferred making brow raises (AUs 1+2) and winks (AU 45) significantly more often on the right and AU 2 tended in this direction.

To test the significance of AU as a variable in determining asymmetry scores, a one-way repeated measures analysis of variance with AU as the independent variable was performed four times, once for each asymmetry measure. The main effect of AU was was significant for the bilateral asymmetry scores, F(9,180)=3.61, p<.001, and better unilateral control scores, F(9,198)=2.41, p<.02, but not significant for the other two asymmetry scores.

The symmetry of actions did not change with repetitions of the requested actions. For each bilateral action requested by verbal or visual instruction, up to four occurrences (first and last two) were scored. All actions showed the same pattern as the totals in Table 3. Neither the ratio of left asymmetrical actions to total asymmetry or percentage asymmetry changed across repetitions of the action.

Relations among Asymmetry Measures

Evidence relevant to the hypothesis that the four measures of asymmetry reflect common underlying processes included: a) consistency about laterality between measures, b) correlations among measures, c) factor analysis of scores. First, the four measures generally showed the same pattern of laterality for any particular action in Table 2, as predicted. If there was a significant pattern of laterality for an AU for any measure, the other measures tended to show the same laterality, although this tendency may not have been significant. Table 4 summarizes the numbers of right versus left subjects for the actions measured by the four asymmetry measures. Inspection of Table 4 indicates possible inconsistency in laterality between two or more measures for AUs 4 (brow lowered), 9 (nose wrinkle), 12 (smiling), and 20 (lip stretch). To test the significance of these exceptions, McNemar tests showed the differences in the proportion of subjects who changed from left to right versus right to left. Table 5 lists the pairs of different asymmetry measures that had significant shifts in these proportions. For four of the five shifts, the direction was towards more right subjects in the ratings of difficulty or preferred side scores versus the other two measures.

Table 6 shows the Pearson correlations between continuous scores of the four asymmetry measures for each AU. Of the 60 correlations in Table 6, 31 are significant, and all significant correlations are positive. No single pattern of intercorrelations seems to apply to all the AUs in Table 6. Table 7 shows the number of significant correlations between each of the different asymmetry measures and the total number of significant correlations for each measure. Better unilateral scores have the largest number of significant intercorrelations with the other measures, and bilateral asymmetry scores have the fewest number of significant intercorrelations. Two possible explanation could account for this difference. It could be an artifact of the measures (e.g., unreliability, restricted ranges), or it could indicate differences between bilateral and unilateral movements since the measures of bilateral movements in Table 7 are the only measures that did not involve unilateral activity. This fact suggested that the unilateral measures tap specific variance different from the bilateral asymmetry measure. Factor analysis was used to examine this latter possibility.

With only 33 subjects, all 40 asymmetry scores for the individual actions could not be included in the factor analysis. Instead, scores were averaged across the ten AUs within each asymmetry measure to provide a score for total asymmetry as determined by each method. Since the combined bilateral asymmetry scores alone might not be expected to define a separate factor, the original verbal and visual scores were used in the first analysis. The principle components of these scores were extracted and factors with eigenvalues greater than 1 were rotated with the varimax technique. A second, similar analysis added the scores for actions during simulations of emotion (discussed below) to provide three scores for bilateral movements. Because different AUs have different patterns of asymmetry, only AUs that were scored with the four measures of asymmetry (the first ten actions listed in Table 4) were retained for the averages of the simulated scores.

Table 8 shows the loadings on rotated factors for the two analyses. In both cases, measures of unilateral actions and the ratings of unilateral difficulty loaded higher and together on a factor different from measures of bilateral actions. Notice that the preferred unilateral movements and the ratings were least related to the factors formed by the verbal and visual bilateral asymmetry scores, paralleling the low frequency of significant correlations between the respective variables. These analyses show that the three measures involving unilateral actions formed a large factor that accounted for about 40 percent of the variance in these measures which was distinct from the variance of bilateral measures, another 20 to 30 percent. Bilateral measures did not load on one factor when three scores for bilateral movements were included in the second analysis.

Relationship of Asymmetry for Different AUs

Different AUs, Same MeasureTable 9 shows the significant correlations between scores for different AUs within each asymmetry measure. For each measure there were 45 correlations between different AUs, but only a small percentage of these were significant. In general, the asymmetry of particular AUs show a complex pattern of incorrelations. Significant intercorrelations were not random, but rather tended to occur predictably between certain groups of AUs. The groups of AUs generally were the same across the different asymmetry measures. These results and inspection of the correlation matrices suggested that there might be only a few factors contributing to the intercorrelations of asymmetry scores within each asymmetry measure. The principle components of scores were extracted for each set of asymmetry measures with mean substitution for missing values. The first three factors for each asymmetry measure were rotated with the varimax technique. Table 10 shows the loadings of scores on these rotated factors. These analyses showed that the three factors could account for about 60 percent of the variance of the ten individual scores for AUs. Although inspection of the factor loadings suggested good consistency between the factors for the preferred side and the rated easier scores, there was no apparent congruence between the factors for these two sets of scores and the factors for the other two sets.

Different AUs, Different MeasuresTable 11 shows the significant correlations between scores for different AUs as measured by each pair of the four asymmetry measures. For each pair of asymmetry measures, there were 90 different correlations between different AUs. The percentage of significant correlations was even less here than for the correlations within each asymmetry measure. Again, these intercorrelations typically were predictable, occurring among certain groups of AUs that were the same groups that appeared within each asymmetry measure.

Unilateral Control and Strength of Contraction

More detailed examination of unilateral actions revealed that truly unilateral actions were infrequent (see Appendix Y); there was usually at least a trace of activity on the side of the face that was supposed to show no activity (the offside). For AUs in the lower face, the actions during unilateral requests were usually asymmetrical, with greater intensity on the side that was requested (the onside). For brow actions, however, most subjects could only make one side of their face have the greater intensity and could not reverse the asymmetry. Even so, these actions will continue to be called unilateral actions because they were elicited by requests for unilateral actions and this label will reduce confusion with the other kinds of actions obtained in this study.

In the discussion that follows, the "onside" of a unilateral action is the side that the subject was requested to make the action on. The "offside" is the other side, the one that the subject was not supposed to move (see Appendix Y for details). For example, in a request for a smile on the left side, the left side would be the onside, and the right side, the offside. Table 12 shows the relation between the better unilateral score for each AU and whether the right or left unilateral request had the greater onside or offside intensity. An unequivocal pattern emerges from these figures. The side of the face that is better at making a unilateral action (defined by differences in asymmetry) contracts more intensely when that side is requested than the other side of the face contracts for a requested unilateral action on this poorer side. Conversely, during a unilateral action on the better side, the contraction on the other side is less intense than the contraction that occurs on the unrequested side during a request for unilateral action on the poorer side. To summarize, the onside contraction tends to be greater when unilateral actions are requested on the better side. The offside contraction tends to be greater when unilateral actions on the poorer side are requested.

Table 12 also shows the relation between the better unilateral action and the occurance of other, unrequested actions. For each AU requested, Table 12 shows whether unrequested actions occurred during the request for unilateral action on one side, but not for the request on the other. When no unrequested actions occurred for either unilateral requests or occurred for both unilateral requests, the events were dropped from further consideration. Although too few events remained for each AU singly, the totals show a clear pattern. As predicted, requested unilateral actions on the better side were more likely to be performed without unrequested actions than unilaterals on the poorer side.

Laterality of Spontaneous Actions

Spontaneous Happy Expression

Most subjects smiled when hearing the question "Now that this is done, aren't you glad it's over?" at the end of the emotion simulations. Many of these smiles seemed to be genuine expressions of happiness or relief, but a few seemed tentative and forced as though to say "He's got to be kidding." Since there was no objective way to distinguish how emotional and spontaneous these smiles were, all were retained in the subsequent analysis. The problem of identifying spontaneous emotional movements is common to every eliciting circumstance, but any deliberate, controlled smiles retained would work against the hypothesis of a difference between spontaneous and deliberate actions. The distribution of asymmetry scores for these spontaneous smiles is given in Table 13. There were more AU 12s stronger on the left than the right, but not significantly more. AU 6 was largely symmetrical.

Spontaneous Startle Actions

Table 14 shows the distribution of symmetry scores for each AU and condition that had a startling noise. Two results are apparent: the incidence of asymmetry was low, and, second, the only AU to show lateralization was AU 6, which was stronger on the right.

Laterality of Mixed Deliberate and Spontaneous Actions

Startle Simulations

Table 15 shows the distribution of asymmetry scores for AUs that occurred frequently in the startle simulations. The incidence of asymmetry was low, and there was no significant lateralization. There were no differences in asymmetry between the two simulations.

Simulations of Emotions

Table 16 shows the symmetry categories of subjects for each emotion simulation (happy, sad, fear, anger, surprise, and disgust) and for each AU that was commonly observed in the simulated condition. Few AUs manifested significant laterality, but trends tended to be in the same direction as the asymmetry measures of deliberate actions in Tables 1, 2 and 4. Significant laterality in the number of left verses right subjects for AUs 2 and 12 was the same as the requested bilateral actions. Table 17 shows that the average asymmetry of actions in each emotion simulation was not significantly lateralized.

Table 18 shows the correlations between total scores for each AU in the simulate condition and the scores derived by the different asymmetry measures in Table 2. Simulations significantly correlated with the requested bilateral actions for AUs 1, 2 and 12 and with both better unilaterals and preferred side scores for AUs 9 and 10. This correlation is consistent with the expectation that deliberate actions are mixed with spontaneous actions in simulations.

The laterality of actions did not shift with increasing involvement of AUs in the simulated expression. It was possible that the more subjects involved themselves in simulating the expression, the more facial actions they would show. This involvement, perhaps indicating emotional processes, could be reflected in the symmetry of actions. For each simulation, Table 19 shows the frequency of right, left and symmetrical categories for certain AUs according to the number of other AUs in the simulation. The AUs tabulated in column 1 of Table 19 occurred frequently in different combinations with other AUs and showed variance in asymmetry. The table shows in column 2 what other actions were in the simulation and shows the number of other actions in column 3. The frequencies show no apparent change in the relative number of right versus left asymmetries with changes in the number of other AUs in the expression. However, the totals across these actions show that the proportion of asymmetrical actions apparently declined with increasing involvement of additional AUs. This result is consistent with the hypothesis that greater emotional involvement in simulations is associated with less asymmetry.

If asymmetry is related to emotional versus more deliberate processes as hypothesized by Ekman et al. (1981), then asymmetry of actions in simulations might be related to how much emotion was involved in producing them. Self-reported ratings of emotion experienced in each simulation were used as a measure of emotional involvement, although there was no reliability or validity data on them. These self reports were correlated with the total degree of asymmetry (ignoring whether right or left) summed across individual actions in each simulation. For happy simulations, the more asymmetrical the expression, the less emotion was experienced (r = -.39, p < .03), just as predicted. However, for fear simulations, the relationship was the opposite (r=.54, p < .01), and none of the other correlations were significant, clouding interpretation of these ambiguous results. These findings should be clarified by further research.

 Differences in Asymmetry between Types of Movement

Table 20 shows significant differences in the degree of asymmetry for AUs that appeared in different conditions. Three different types of actions are represented in this experiment: spontaneous emotional or reflex actions, deliberate requested actions, and more ambiguous types of actions occurring when the subjects simulated expressions.

Spontaneous Actions versus Deliberate, Requested Actions

Absolute values of asymmetry scores were used as measures of degree of asymmetry, regardless of whether it was right or left. Wilcoxon signed-ranks tests showed that, as predicted, spontaneous actions were more symmetrical than deliberate actions. AU 12 in spontaneous happy expressions was more symmetrical than in requested actions (z=4.53, p<.001). AU 6 in spontaneous happy expressions was more symmetrical than the requested action of AU 6 (z=2.42, p<.02). AU 20 in the startle expressions was more symmetrical than requested actions of AU 20 (z=3.19, p<.002). AU 7 in startle expressions tended to be more symmetrical than requested actions of AU 7 (z=1.69, p=.09). The prediction was not confirmed for AUs 6, 45, and the eye closure measure between the startle and the requested actions, all showing no difference. AU 6 was unexpectedly more asymmetrical in the startle than in the spontaneous happy expression (z=2.27, p<.03).

These differences in asymmetry between conditions cannot be attributed to differences in the intensity of actions between conditions. In a separate analysis to eliminate intensity differences, each action in the spontaneous conditions was matched for intensity with a deliberate action since there was a range of intensity scores among the numerous deliberate actions. This analysis showed the same differences as the one based on averages of all actions shown.

Differences between Startle Conditions

Few differences between startle conditions where subjects heard a noise were found. Absolute values of asymmetry scores were used to explore differences in the degree of asymmetry regardless of whether the asymmetry was right or left. These values were averaged across AUs in each condition to provide a measure of total degree of asymmetry in each startle. Wilcoxon signed-ranks tests on these averages did not show any significantly different pairs of conditions. Signed-ranks tests on each AU separately showed a difference only for the eye closure measure between the unanticipated and anticipated condition (p < .05). Although the anticipated condition showed greater asymmetry as predicted, this difference could be attributed to chance since no other prediction about differences between conditions was confirmed.

Signed-ranks tests on the original asymmetry scores and on the averages of these scores were used to determine whether the amount of right or left asymmetry differed between conditions. No differences in the averages of asymmetry scores were found between conditions in Table 14. Looking at each AU individually showed that AU 7 shifted to the right in the inhibit condition from left in the unanticipated condition (p < .05). This result can easily be attributed to chance since there were 29 other comparisons that showed no significant differences.

Spontaneous Startle versus Startle Simulations

The same kinds of analyses as above were conducted for the two simulated startles versus the startles in response to a noise. The two simulations did not differ significantly in the actions that subjects produced, but the simulations had more actions that were not common in startles (e.g., AUs 1, 2, 12) and fewer actions that were common in startles (e.g., AUs 6, 7, 20). The only difference in asymmetry between simulations and real startles can be attributed to chance.

One reason why simulations did not differ from startles in asymmetry might be that the pencil tap was conducive to recruiting a partial startle. The sound pressure level of the tap was far too low to elicit a genuine startle, but the tap had some characteristics, such as a sharp onset, that might have helped subjects to simulate more than other alternatives, such as an experimenter's verbal prompt.

Mixed versus Spontaneous or Deliberate Actions

There were few significant differences in the degree of asymmetry between the mixed types of actions in the simulated expressions and either the spontaneous or deliberate actions. Spontaneous actions of AU 12 were more symmetrical than AU 12s in emotion simulations of happy (z=2.92, p<.004), which were, in turn, more symmetrical than deliberate actions of AU 12 (z = 6.66, p < .001). AU 20 in the startle tended to be more symmetrical than AU 20 in the emotion simulations (z=1.75, p<.08). As predicted, the degree of asymmetry of simulated actions was intermediate between spontaneous actions and the requested, deliberate actions, even when differences were not significant. Table 21 lists each subject's raw scores for AU 12 in three types of movement, deliberate (bilateral asymmetry), simulated, and spontaneous emotional.

Symmetry and Personality

Considerable effort was devoted to assessing the relation between asymmetry and personality. Some researchers have reported a relation between facial expression and repression (Notarius & Levenson, 1979), and Schwartz (personal communication) suggested that this variable may mediate lateralization of asymmetry in facial actions. Repressors were defined as subjects who scored below the median on the Taylor Manifest Anxiety Scale and above the median on the Social Desireablility Scale. True low anxious subjects were defined as those who scored below the median on the Anxiety Scale and below the median on social desireability (Weinberger, Schwartz, & Davidson, 1979). There were seven true low anxious subjects and eleven repressors so identified. The remaining subjects were classified as moderately anxious subjects. Many analyses were performed to detect differences between these groups, but they were virtually identical on all the asymmetry measures used in this study. Inspection of the distributions of left, right, and symmetrical categories of subjects showed no consistent differences. This lack of relation assures that the other findings of this study are not mediated by the repression variable.

The relations between personality variables and asymmetry were explored post hoc with correlational techniques. These analyses produced little evidence that symmetry is related to any of the personality variables used in this study. Only about six percent of these correlations were significant, no more than might be expected by chance. Significant correlations that did occur were distributed equally across measures of asymmetry and personality with two isolated exceptions which, since they were not predicted, should not be interpreted until replicated. First, significant correlations between asymmetry scores and personality measures were disproportionately high for brow raises (AU 1+2) (Chi-Sq. = 12.7, p < .001). Second, the "ego-organization" scale of the Adjective Self-Rating Questionaire (Veldman & Parker, 1970) was related to asymmetry of brow raising (AU 1+2) for three of the four asymmetry measures (bilateral asymmetry, r = -.34, p = .05; better unilateral, r = -.35, p =.05; preferred side, r = -.37, p<.05). Further research on the relation between asymmetry and personality might profitably concentrate limited resources on measuring asymmetry in the brow area, ego-organization personality variables, and anxiety variables which were negatively related to ego-organization and showed scattered significant correlations with asymmetry.

Laterality within Subjects across AUs

Thus far, laterality has been examined in terms of the consistency in asymmetry across subjects for each Action Unit. Another way to look at the data from this study is to examine consistency within each subject across the actions measured. This approach addresses the issue of whether any particular subject has a bias for one side of her face, and, if so, is it the same side for different subjects. Table 22 shows the significant laterality manifested by individual subjects across AUs for each asymmetry measure. A maximum of eight actions was measured by the verbal bilateral asymmetry score, eleven by the visual bilateral asymmetry score, and ten by each of the other three measures. The cell entries indicate significant laterality for a given subject and score. Laterality was defined by a generous criterion of a difference between the number of right and left actions that had less than a one in ten chance of occurring randomly according to a two-tailed binomial test. A more stringent criterion would have eliminated all but a few subjects from the analysis.

Inspection of Table 22 shows that few subjects manifested laterality. This result cannot be attributed merely to the criterion for laterality or to infrequent asymmetry because most subjects showed a mixture of left and right asymmetries for each measure. Except for the better unilateral control score, there is slightly more right laterality in Table 22, but not significantly more for any measure except preferred side which showed six right subjects and no left (p<.05, two-tailed binomial). In contrast to laterality across actions, the rated easier scores showed the most laterality while bilateral asymmetry scores showed the least.