We thank Peggy Thoits for her useful comments on an earlier draft of this paper. We also thank participants in social psychology seminars at Indiana University's Department of Sociology for critiques and suggestions.

Abstract

Affect Control Theory (ACT) posits that emotions transform people from their general identities into the immediate situated identities that are emerging in events. The perspective is formalized in a model that predicts emotions. We tested ACT predictions of emotions in 128 events against self-reported emotions of respondents imagining themselves in the happenings. Results indicate that people rarely report emotions far from the ACT predictions, and that self-reported emotions usually are close to the ideal emotional states predicted by ACT. However, some emotional states close to the ACT ideal may be chosen rarely, indicating that cognitive factors in addition to ACT contribute to selection of emotions. Overall, ACT predicted self-reported emotions in about 80 percent of the events considered.

Predicting qualitative features of social interaction, such as what behaviors will occur or what emotions will arise, seems too much to ask of social science. We do not expect sciences to predict actual details about natural circumstances. For example, we put our lives in the hands of aeronautical scientists who design aircraft, without expecting that those scientists are able to predict paths of falling leaves.

Yet one framework within sociological social psychology does concretely specify possible happenings in a social interaction, given a verbal definition of the situation. Affect Control Theory (e.g., see Heise, 1977, 1979; Smith-Lovin, 1979; Smith-Lovin and Heise, 1988; MacKinnon, 1994; Smith, Matsuno, and Umino,1994) takes a naturalistic orientation and sophistication about how minds work from symbolic interactionism. The theory's predictive potential comes from survey data on cultural sentiments, employed in a mathematical model using empirically-derived equations.

Affect Control Theory (ACT) posits that a person's standard orientation and demeanor are determined by an identity that sets the person's status (how worthy or unworthy the person is), power (how potent or impotent the person is), and expressiveness (how spontaneous or reserved the person is). Predictions about social interaction are obtained by specifying participants' identities in a computer program called Interact. The program retrieves cultural sentiments for the identities, computes the general kinds of behavior that would best confirm the identities, and translates its quantitative results into words.

For example, defining one person as Physician and the other as Patient leads to behavior specifications that have the Physician "listening to", "cautioning", and "medicating" the Patient, while the Patient "observes" and "submits to" the Physician. These behaviors match Interact's quantitative specifications that both interactants should act friendly and serious to confirm their identities, but the potent Physician should act in a capable and commanding manner, while the Patient should behave submissively.

ACT behavior predictions usually have face validity in the sense that they include obvious role expectations for each identity. Additionally, a number of studies have empirically validated the theory's formulations regarding behaviors.

A laboratory experiment by Wiggins revealed that ACT's predictions about general kinds of behavior correlate with observers' ratings of actual interaction. Additionally, the experiment confronted gladdened and demeaned subjects with high and low status others. Gladdened subjects acted friendly to all, but demeaned subjects were observed ingratiating themselves with high-status others and avoiding low-status others, as ACT predicts (Wiggins and Heise,1987).

Respondents in a survey gave ratings of unlikely to events that disconfirmed identities of interactants (Heise and MacKinnon, 1987). The unlikely ratings were given to events that made people look too good, as well as to events that created excessively bad impressions of participants, reinforcing ACT's assumption that any kind of disconfirming event is exceptional because people behave so as to confirm identities.

Members of a gay Christian church had positive sentiments about gay identities, so ACT predictions were that they would act friendly and supportively with other gays and with religious leaders (Smith-Lovin and Douglas, 1992). Field observations supported ACT predictions, and demonstrated that gays did not act anxious and defensive, as predicted from sentiments about gays held in another church. Interviews with gay church members revealed that they thought events predicted from their own sentiments were more likely than events predicted from sentiments of non-gays or events created with random behaviors.

A laboratory experiment showed that people with low self esteem seek negative appraisals from others, even though negative appraisals make these people feel bad, and positive appraisals make them feel good (Robinson and Smith-Lovin, 1992). Subjects with low self esteem feel their critics are accurate, competent, and attractive, and they choose critics rather than admirers as future interaction partners. On the other hand, subjects with high self esteem prefer the appraisals and company of admirers. All of these results correspond to ACT formulations about behavior.

Thus research shows that ACT's predictions about interpersonal behaviors do correlate with what people expect, and the theory correctly predicts general kinds of action from an operative definition of the situation. Additionally, ACT correctly predicts specific behaviors that people might be motivated to do.

Predicting Emotions

ACT also predicts emotions that should arise in social interaction. A person's identity defines the kind of presence he or she ideally should have. Specific events modify the person's apparent worthiness, power, and spontaneity, in ways that are described by impression-formation equations (Heise, 1979; Smith-Lovin, 1987a). Thus at any moment the person has two standards defining the worthiness, power, and spontaneity that he or she should manifest. Emotion acknowledges both.

ACT's predicted emotion combines with the interactant's identity and creates the same impression as is being created about the interactant within the current event. In essence, ACT posits that emotions transform people from their general identities into the immediate situated identities that are emerging in events. An emotion makes a person in a given role feel and look the way events have made that person seem.

For example, ACT predicts that a Physician and a Patient interacting in the ways specified above both will feel at ease or feel no emotion at all, but the Physician will be somewhat disposed to mild positive feelings like contentment, while the Patient will be more disposed to melancholy. On the other hand, a physician failing to keep an emergency-room patient alive would experience fright, adjusting his ordinary high status, power, and reserve so that he personally manifests the unworthiness, impotency, and hyperactivity of failing to prevent a great misfortune.

ACT's emotion model is derived from empirical equations describing the impressions that result from combining modifiers and identities (Averett and Heise, 1987; Heise and Thomas, 1989). The meaning of a modifier-identity combination, like "angry father," is influenced by each of the constituent meanings. Solving for emotion creates emotion attribution equations defining the emotion that modifies a given identity into a given impression. The derived emotion, defined in terms of its levels of pleasure, dominance, and activation, answers the question of what emotion should be felt and displayed, given a specific identity and a specific impression created by events.

ACT's formulation about emotions is distinct from its formulation about behavior. Behaviors are selected so that impressions created by future events will match identities as much as possible. Emotions are selected so as to reflect the identities and impressions that actually exist at a given moment. Thus, empirical support for the behavior model does not validate the emotion model directly.

ACT's emotion predictions often have face validity. However, sometimes the predictions are non-intuitive (e.g., when predicting depression rather than anger over others' bad deeds), which undercuts face validity, even though making the predictions more intriguing. In any case, face validity possibly varies among observers, so empirical tests of predictions are required to assess the model objectively.

Experimental studies of how emotions influence legal outcomes (Smith-Lovin and Tsoudis, 1993; Robinson, Smith-Lovin, and Tsoudis, 1994) provide one kind of validation of ACT's emotion model. The studies show that prison sentences given to convicted defendents are less punitive when the defendents display an appropriate emotion like remorse over illegal deeds. An actor who shows remorse retains her claim to a high-status identity while acknowledging that her act was bad. On the other hand, showing glee or pride when associated with a deviant act suggests that the actor thinks of herself as the kind of person for whom such action is self-confirming, thereby encouraging others to label the actor as a deviant. These experimental results accord with a derivation from ACT's emotion model (Heise, 1989), that inappropriate emoting results in stigma.

However, no study has compared ACT predictions about emotions with people's actual reactions. We do not know yet whether ACT predicts specific feelings that people have in social interaction.

Testing the Emotion Model

The study reported here compares ACT emotion predictions with the emotions that people say they feel while imagining various social situations.

Data were collected as follows. A respondent read a hypothetical event and identified the emotion he or she felt while imagining that circumstance. Respondents chose from 25 emotions. The options, constituting about one-fourth of the emotion words in English, reflected the full range of emotional states that can be named. The use of words to describe emotions fits Averill's (1982, p. 329) directive that "any theory of emotion must in the end relate to the kinds of phenomena, no matter how complex, that are recognized as emotions in ordinary language."

The respondent's identity in the event was set theoretically as "I, myself". In half of the 128 events, the respondent was the actor in the event, in the rest of the events the respondent was the object of action. The kind of event was varied by changing the action and by changing the status, power, expressiveness, and institutional affiliation of the other person in the event.

Theoretical predictions were obtained by computing the impression created of the focal person in each event, using impression formation equations (Heise, 1979; Smith-Lovin, 1987a) and average ratings acquired in past studies. A computed impression is a profile of numbers indexing how worthy, powerful, and spontaneous the person seems to be as a result of the event. Or, in the language of affect control theory, the computed impression is an Evaluation-Potency-Activity (EPA) numerical profile. (Evaluation, Potency, and Activity are universal dimensions of affective meaning - Osgood, May, and Miron, 1975.) The impression created by an event, plus the EPA profile for the identity that the respondent had in the event, allow the theoretical emotion to be predicted with the emotion attribution equations.

A social happening usually evokes a diffuse emotion that is named differently by different people (Heise and Calhan, 1995). Thus a theory of emotion should predict not a single emotion but rather should bracket emotions that are likely to be identified in a given happening. In our tests, we order emotions for a given happening according to how close they are to a prediction. The general hypothesis is that people will choose the closest emotions frequently, while the emotions that are farthest from the prediction will be chosen rarely or not at all. "Closeness" was assessed as the Euclidian distance between the EPA profile for the predicted emotion and the EPA profile for an emotion option chosen by respondents.

Procedures

Part one of this study uses data on emotional reactions reported by Heise and Calhan (1995). Additional details on selection of stimuli and on instrumentation can be found in that article. Part two of this study focuses on four stimuli selected from part one, with greatly enlarged samples of respondents.

Stimuli

Each stimulus described the respondent as engaged with an alter named by a social identity, and either the respondent or alter was performing a social behavior on the other. Respondents were asked to specify the emotion they felt while imagining the encounter. For example, "Imagine a flight attendant is assisting you. How do you feel at that moment?" or "Imagine you're rescuing a hero. How do you feel at that moment?" (Slightly different phrasing was used with nine instances of insinuated behaviors ‹ e.g., "You realize your landlord is evading you. How do you feel at that moment?")

Stimulus events systematically combined Evaluation-Potency Activity configurations for behavior and for the alter identity. The institutional affiliation of alter (academic, business, justice, medicine, religion, laity, family, or intimacy) also was varied systematically in an 8x8 Graeco-Latin square design. Sixty-four scenarios presented the respondent as acting on alter. Another 64 scenarios described the respondent as the object of alter's action. All 128 events are listed by Heise and Calhan (1995).

The 128 stimuli were distributed into two questionnaires, such that no identity-behavior combination appeared more than once in a form, and respondent-as-object and respondent-as-actor were represented equally in each form.

Gender comparisons are based on the sex of the respondents, determined through an item on the questionnaires.

Instrumentation

Emotion options were configured graphically as a spiral, which allowed respondents to select rapidly from 25 specific emotional states: annoyed, ashamed, at-ease, bitter, calm, contented, depressed, disgusted, embarrassed, excited, flustered, furious, happy, impatient, joyless, mad, nervous, no-emotion, outraged, overjoyed, pleased, proud, scared, thrilled, and unhappy. Pleasant feelings appeared at the top of the spiral, activated feelings at the right, and overlapping sectors of the spiral showed vulnerable feelings on the inside and dominant feelings on the outside. Additionally, check boxes allowed respondents to record emotional intensity and to select "no emotion" as an answer. A picture of the instrument and some details on its usage are provided by Heise and Calhan (1995).

Data Collection

To obtain data for part one of the study, questionnaires were distributed in two sociology classes. Of the 132 respondents who returned questionnaires, 125 provided usable data for all analyses, or about 62 respondents for each of the two forms. Data for part two of the study were obtained in eight other sociology classes; 617 respondents returned questionnaires, and 610 provided usable data on two or more stimuli.

The questionnaire administrator orally interpreted the instruction sheet attached to the questionnaire, and respondents completed the questionnaire in class. The group administration took about 25 minutes for part-one questionnaires and about 10 minutes for part-two questionnaires.

Results

Part 1

Each event presented to respondents provides data on how often males and females chose each of 25 emotions to identify their feelings in a specific happening. The proportions choosing various emotions should correlate inversely with distances between the emotions and the feeling predicted by the theoretical model. In other words, emotions that are close to a theoretical prediction should be chosen often.

Table 1 summarizes the results over all 128 events. In the first part of the table, Spearman's rank-order correlation coefficient (Rho) is used as a measure of association between proportions and distances. Ideally, the rank order correlation for each event should be negative, indicating that small distances predicted large proportions and large distances predicted low proportions. The table shows percentages of events in which the correlations were negative and also shows the median values of the rank order correlations.

Table 1 Percent of 128 Events With a Negative Correlation Between Proportions of Respondents Selecting Emotions and Distances of Emotions From Model's Prediction, by Sex of Respondent
	Femalesa	Malesb
Percent with negative Rho	88*	84*
Median value of Rho	-0.32	-0.29
Percent with negative Mu2	89*	81*
Median value of Mu2	-0.61	-0.62
a Maximum respondents per event: 41. b Maximum respondents per event: 22. * Binomial probability less than .05 if probability of a negative correlation is 0.50, one-tail test.

ACT emotion predictions were successful, in the sense that negative correlations occurred instead of positive correlations, in more than 80 percent of the events considered, for both females and males. Such results are well beyond what would be expected for a null hypothesis where the average rank order correlation is zero. Thus, the theoretical emotions defined by affect control theory are predictive of the emotions that people feel in imagined events.

Median values of the rank-order correlations are -0.32 for females and -0.29 for males. We examined scatter plots of predictions versus frequencies to see why absolute magnitudes of the rank order correlations were not greater, and we found that the distributions typically are asymmetric, fitting the following generalizations.
(1)The emotion that a person reports feeling in an event usually is close to the theoretical emotion predicted by affect control theory.
(2)People rarely report feeling an emotion in an event that is far from the theoretical emotion predicted by affect control theory.
(3)However, some emotions that are close to the predicted emotion rarely are chosen by any one.
The first two propositions describe a linear relation, going from high proportions of choice for small distances, to low proportions of choice for large distances. The third proposition indicates that some emotions fall beneath the line, such that low proportions of choice occur with small distances. Thus the relation between proportions of choice and distances from the predicted emotion is triangular rather than linear.

Guttman's Mu2 coefficient (Shye, 1978, Appendix) measures fit to a triangular pattern better than the rank order correlation, so we repeated the analyses using Mu2 instead of Rho as a measure of relation between proportions and distances. The bottom two rows of Table 1 summarize these results. Using Mu2 as a criterion, affect control theory still operates successfully in more than 80 percent of events for both females and males. However, the median association between frequencies and distances is twice as large when measured by Mu2 instead of Rho, reflecting the triangular shape of the relation, as described in the propositions above.

Table 2 Percent of Events With a Negative Mu2 Coefficient, by Sex of Respondent, and by Agency of Respondent in the Event
	Femalesa	Malesb
64 Events With Respondent as Agent
Percent with negative Mu2	81*	73*
Median value of Mu2	-0.56	-0.56
64 Events With Respondent as Object
Percent with negative Mu2	97*	89*
Median value of Mu2	-0.66	-0.62
a Maximum respondents per event: 41. b Maximum respondents per event: 22. * Binomial probability less than .05 if probability of a negative correlation is 0.50, one-tail test.

Table 2 presents results just for events in which the respondent was the agent of action, or in which the respondent was the object of action. The table shows that affect control theory predicts actor emotions and object-person emotions well for both sexes. However, the theory does a somewhat better job of predicting object-person emotions than actor emotions, and of predicting the emotions of females than of males.

Part 2

Only small samples of respondents were available to examine each event in Part 1 of this study. Emotions with moderate probabilities of being chosen and emotions with very low probabilities both would have near-zero proportions of choice in small samples, which could create a triangular pattern artifactually. Clarifying whether the relation between proportions and distances is linear or triangular requires a sample of respondents that is large enough to eliminate "floor" effects.

Proportions are estimated with error from a small sample. This could cause the median correlations reported in Tables 1 and 2 to be too low, inasmuch as errors in measurement generally cause attenuation of correlations. Estimating the actual degree of association between choice proportions and distances requires larger samples of respondents.

Part 2 of this study employs relatively large samples of respondents while focusing on four selected events. The results provide information about the impact of sample size on the relation between proportions and distances. Additionally Part 2 offers a closer look at several events in which emotions were not predicted correctly by affect control theory in Part 1.

Table 3 gives an overview of results from Part 2, and also shows parallel information from analyses in Part 1 for the sake of comparison.

Table 3 Median Rho's and Mu2's Measuring Association Between Proportions of Respondents Selecting Emotions and Distances of Emotions From Predictions for Four Events
	Small Sample						Large Samplea
	Females			Males			Females			Males
	Rho	Mu 2	N	Rho	Mu 2	N	Rho	Mu 2	N	Rho	Mu 2	N
A library assistant is neglecting you	-0.18	-0.67	40	-0.28	-0.67	20	-0.19	-0.72	386	-0.40	-0.74	221
A handicapped person is helping you	-0.57	-0.65	39	-0.21	+0.08	17	-0.61	-0.89	385	-0.55	-0.78	221
You're silencing an expert	-0.03	-0.29	34	+0.07	0.00	18	-0.50	-0.63	357	+0.15	+0.25	217
You're seducing an innocent	+0.10	+0.39	36	+0.17	+0.12	17	-0.12	+0.36	335	+0.05	+0.01	209
a The large samples do not include respondents in the small samples.

The first event, A library assistant is neglecting you, was selected as a typical case in which the coefficients of association obtained in Part 1 are close to the median values across all events in Part 1. The left side of the table shows the Rho's and Mu2's for this event as obtained in Part 1 of the study, along with the sizes of the female and male samples. Rho's and Mu2's obtained in Part 2, with samples about ten times bigger, are given on the right side of the table.

Even with larger samples, the absolute magnitude of Mu2 continues to be substantially bigger than the absolute magnitude of Rho, suggesting that the triangular relation is sustained. Indeed, one emotion, annoyed, is chosen very frequently (60% females, 40% males), whereas other emotions predicted by affect control theory are chosen infrequently. Meanwhile, emotions that are predicted to be irrelevant are chosen infrequently, by both females and males. Thus, the triangular relation persists.

The absolute magnitudes of Mu2 increased in the larger samples, suggesting that the coefficients in Part 1 are attenuated due to error in estimating proportions. However, the increases in magnitude are modest for this event -- 0.05 for females and 0.07 for males.

The second event, A handicapped person is helping you, was chosen for inclusion in Part 2 because it yielded typically successful results for females in Part 1 but not for males. Further study of this event might clarify why affect control theory predictions are somewhat better for females than for males.

With large samples, the female and male results are comparable, successful predictions being made in both cases. This suggests that the superiority of predictions for females in Part 1 arose at least partly from imperfect estimation of male proportions as a result of small samples of males in Part 1.

Rho's and Mu2's both are moderately negative in the large sample results for this event, suggesting that the triangular relation is less pronounced. In this particular event, emotions closest to the theoretical predictions have moderate proportions of females and males choosing them, so Rho achieves a larger magnitude than usual.

Once again, the absolute magnitudes of Mu2 increased in the larger samples, suggesting that the coefficients in Part 1 are attenuated due to error in estimating proportions.The increases in magnitude are substantial for this event -- 0.24 for females and 0.86 for males.

The third event, You're silencing an expert, was chosen to check again the relative superiority of predictions for females, this time with an event where the success for females was modest in Part 1, while the male data showed no relation between proportions of respondents selecting emotions and distances of emotions from a model's prediction.

The large sample produced substantial increases in the magnitude of correlation coefficients for females -- 0.47 for Rho and 0.34 for Mu2. The large sample of males also increased magnitudes of coefficients, but in the direction of more positivity. That is, with this event, large samples increased the clarity of affect control theory's successful predictions for females and also increased the clarity of the theory's failure to predict male responses. Examining detailed results reveals that the triangle pattern appears again in the female data. Moreover, a fairly good triangle also is present in the male data, except for one outlier -- a high proportion of males choosing proud. Overall, the theoretical predictions for males would have corresponded better to data if ACT had predicted more emotional pleasure for males.

The fourth event, You're seducing an innocent, was selected as an event for which affect control theory clearly failed to predict emotions in Part 1. The large samples in Part 2 yield the same result. Examining the proportions of respondents who chose each emotion reveals that the usual triangle relation actually is obtained for both females and males, with the exception of one outlier - the emotion ashamed. Shame should not be felt according to affect control theory, but it is reported by substantial proportions of both females and males. This result is examined in the Discussion.

Discussion

Research on the sociology of emotions has yielded insight into how affect functions in social life plus theories about how specific emotions emerge in social interaction (see reviews by Thoits, 1989; Kemper, 1990a; Smith-Lovin, 1994). Affect control theory (ACT) addresses both considerations, with its focus on affective sustainment of sociocultural structures, and its predictions of emotional responses in verbally-defined happenings. In the research reported here we tested ACT's ability to predict specific emotions that people feel in events.

Our empirical results indicate that people rarely feel emotions that ACT identifies as unlikely in an event. Moreover, the emotions that people feel generally are among those predicted as likely by ACT. This pattern held to some degree in most of the events considered, for agents of action as well as objects of action. Still, the model rarely predicted proportions perfectly, and it predicted proportions erroneously for a few events. We now consider reasons why.

The Expressive Order

Maintaining an expressive order precludes a variety of happenings, but expressive order is only one factor involved in the creation of human activity. Goffman (1967, p. 9) defined expressive order as follows.

By entering a situation in which he is given a face to maintain, a person takes on the responsibility of standing guard over the flow of events as they pass before him. He must ensure that a particular expressive order is sustained -- an order that regulates the flow of events, large or small, so that anything that appears to be expressed by them will be consistent with his face.

This reasoning applies to emotions. Inappropriate emotions all must be avoided to maintain the expressive order, but expressively appropriate emotions generally include options. To illustrate, a person being neglected by a library assistant would disturb the expressive order by looking proud, overjoyed, happy, thrilled, excited, or pleased. On the other hand, several emotions could be expressively appropriate in the situation: flustered, joyless, disgusted, embarrassed, nervous, unhappy, annoyed. Explaining occurrence of one of the expressively appropriate emotions, and in particular why so many respondents feel annoyed, is a job for constructionist (e.g., Averill, 1982; Schieffelin, 1983) and cognitive (e.g., Ortony, Clore, and Collins,1988; Stein, Trabasso, and Liwag, 1993) theories of emotion.

Thus, ACT identifies non-occurring emotions that would disrupt expressive order while predicting more variations in appropriate emotions than actually occur. The case of emotions parallels ACT's performance in other areas. For example, expressively inappropriate events as defined by ACT are rated as unlikely by respondents; but expressively appropriate events include some happenings that people rate as likely and others they rate as unlikely (Heise and MacKinnon, 1987). ACT's predictions of labelings to account for deviant acts typically include specifications that are sensible along with others that seem illogical in a particular situation (Smith-Lovin, 1987b). Overall, expressive order establishes social boundaries while being but one of multiple factors involved in the production of social phenomena, and ACT predictions reflect this fact.

Shared Culture

A second reason why our models dealing with affective meanings failed to predict emotions perfectly is that we relied on the assumption of shared culture.

ACT's predictions about emotions are made from cultural data in the form of measured sentiments about people and social actions. In this study we assumed that each undergraduate respondent shared cultural sentiments assessed from ratings by other undergraduates. For example, each female respondent supposedly felt that she herself is extremely nice, fairly potent, and quite lively; that a handicapped person is slightly good, slightly powerless, and slightly inactive; and that helping someone is quite good, fairly potent, and neither active nor inactive. From these specifications ACT predicts which emotions are likely while a handicapped person is helping a female respondent.

The overall successfulness of ACT predictions justifies the shared culture assumption and accentuates the durability and pervasiveness of culture. We were able to predict respondents' emotions ten years after measuring average self sentiments in the same population, and 18 years after assessing affective meanings of social identities and behaviors at a different geographic location in the U.S. Nevertheless, the shared-culture assumption most likely is responsible for the worst errors in predicting emotions. Affective meanings of some social identities and behaviors vary across sub-cultures and structurally-defined aggregates (Thomas and Heise, 1995). Therefore assuming that everyone has the same sentiments produces some errors when predicting the emotions of individuals.

For example, affective-meaning models did a poor job of predicting the proportions of people feeling various emotions while seducing an innocent. Apparently the act of seducing is interpreted as courting by some respondents who report they feel excitement or nervousness, and these were responses that ACT predicted. Other respondents apparently interpret seduction as a kind of humiliation, and several respondents saw the event so negatively that they refused to answer, writing they would never do it, while other respondents skipped the item without comment. Of those who recorded an emotion, substantial proportions of both females and males reported feeling ashamed, even though shame should not be felt according to ACT predictions. Thus, sentiments regarding seduction vary too extremely to allow predictions of emotions from a single set of sentiment measures. Each respondent would have to be linked to one of several alternative sentiment measurements in order to make accurate predictions.

Self sentiments have a cultural aspect in that average self ratings are similar across geographic regions and over decades of time. Nevertheless, our assumptions in this study that all females have the same self sentiment and that all males share a single self sentiment surely were somewhat inaccurate. To the extent that individuals' self sentiments differed from the averages, reported emotions would be more variable than predicted, because distinctive self sentiments lead to different impressions of self in an event, and also provide different reference points for assessing impressions of self. Variations in self sentiments possibly account for the tiny proportions of respondents reporting all kinds of emotions in events, even some emotions that seem bizarre.

Statistical Considerations

A third reason why theoretical models dealing with affective meanings predict emotions imperfectly is that data on emotions are fallible. In Part 1 of this study we chose to examine interpersonal emotions in a wide range of events, though that meant having few respondents for each event, providing only rough indications of proportions choosing each emotion. The consequent errors in measuring proportions limited how well theories could perform. We examined respondent samples ten times bigger for a few select events in Part 2 of this study in order to assess the seriousness of the limitation.

Correlations between proportions of respondents selecting emotions and distances of emotions from ACT's prediction continued to be negative in Part 2 if they were negative in Part 1, and the correlations grew in magnitude when proportions were measured better. The size of Spearman rank order correlations increased by 0.20 on the average, and Guttman Mu2 coefficients increased by 0.18 on the average. Therefore the Part 2 results suggest that median correlations computed from Part 1 data, as reported in Tables 1 and 2, substantially understate the strength of the relation between ACT predictions and outcomes. Had respondent samples been 200 or more in Part 1, the median Mu2 might have been around -0.80 for both females and males, which probably is about as sizable as one could expect, given inaccuracies introduced by the shared culture assumption.

In Part 1 ACT predictions seemed to be successful slightly more often for females than for males. In Part 2 we examined two events with a negative correlation for females but not for males. In one instance a small positive Mu2 correlation for males turned strongly negative when the number of male respondents was increased. In the other instance non negative correlations for males stayed non-negative. These results suggest that the sex difference in Part 1 arose partly because the small samples of males in Part 1 yielded erroneous estimates of proportions. However, in a few events (mainly those in which the respondent is the agent of action) ACT predicts female emotions better than male emotions. This sex difference, while providing an opportunity for future investigations, needs to be kept in perspective: ACT does quite well in predicting emotions for both females and males.

Our samples of college student respondents, while unrepresentative of any national population, do typify highly literate middle-class young adults in the U.S., and their responses have allowed us to test predictions of emotions from affective meanings in at least one fairly homogeneous culture. The theory would have lost viability had it not succeeded within this important sub-group of U.S. society. Beyond that, success in one culture provides hope that success will be attained in other cultures as well ‹ a hope that is bolstered by ACT research on the ubiquity of impression-formation processes cross-culturally (Smith-Lovin, 1987a; Smith, Matsuno, and Umino, 1994).

Conclusion

Lewis and Haviland's (1993) Handbook of Emotions offers a plethora of theories focusing on physiological, psychological, and social aspects of affect. Most are explanatory accounts of emotion processes rather than attempts to predict which qualitative emotion will be felt in specified circumstances. Kemper's theory (1978; 1990b) addresses the predictive problem more seriously than most other frameworks, and its predictions have been validated empirically (Kemper, 1991). However, even Kemper's theory requires a great deal of interpretation before a prediction can be made. That is, one must decide whether a person's status and power have increased or decreased in a given situation and who is responsible for the change. Such determinations require expert judgments.

Affect control theory's predictions of emotions require specification of people's identities and of the event in which they are involved. Then previously assembled cultural data on sentiments are employed in empirically derived impression-formation equations, yielding concrete predictions about which emotions might be felt and which will not be felt. Procedural objectivity provides a straightforward connection between emotion issues and a theoretical framework that also predicts normative expectations, behaviors, attributions, and labelings.

This study dealt with emotions that people have when imagining social encounters. It remains to be seen whether emotions reported in imagined encounters parallel lived emotions in actual encounters. Nevertheless, predicting emotions in imagined encounters is an important advance for social science. It means we can understand the emotionality that arises as people mentally plan, rehearse, and fathom social scenes.

References

• Averett, C. P. and D. R. Heise 1987. Modified social identities: Amalgamations, Attributions, and Emotions. Journal of Mathematical Sociology, 13: 103-132.
  
• Averill, James R., 1982. Anger and Aggression: An Essay on Emotion. New York: Springer-Verlag.
  
• Goffman, Erving 1967. Interaction Ritual: Essays on Face-to-Face Behavior. Garden City, NY: Anchor Books.
  
• Heise, D.R., 1977. Social action as the control of affect. Behavioral Science, 22: 163-177.
  
• Heise, D. R. 1979. Understanding Events: Affect and the Construction of Social Action. New York: Cambridge University Press.
  
• Heise, David R. 1989. Effects of emotion displays on social identification. Social Psychology Quarterly, 52: 10-21.
  
• Heise, D. R., and Calhan, Cassandra, 1995. Emotion norms in interpersonal events. Social Psychology Quarterly, 58: 223-240.
  
• Heise and Lewis, Elsa. 1988. Introduction to INTERACT. Durham, NC: National Collegiate Software Clearinghouse, Duke University Press.
  
• Heise and MacKinnon, N. 1987. Affective bases of likelihood judgments. Journal of Mathematical Sociology, 13: 133-151.
  
• Heise, D. R., and L. Thomas 1989. Predicting impressions created by combinations of emotion and social identity. Social Psychology Quarterly, 52: 141-148.
  
• Kemper, T. D. 1978. A Social Interactional Theory of Emotion. New York: Wiley.
  
• Kemper, T. D. 1990a. Themes and variations in the sociology of emotions. Chapter 1 in T. D. Kemper (Ed.), Research Agendas in the Sociology of Emotions. Albany: State University of New York Press.
  
• Kemper, T. D. 1990b. Social relations and emotions: A structural approach. Chapter 8 in T. D. Kemper (Ed.), Research Agendas in the Sociology of Emotions. Albany: State University of New York Press.
• Kemper, T. D. 1991. Predicting emotions from social relations. Social Psychology Quarterly, 54: 330-342.
  
• Lewis, Michael, and Jeanette M. Haviland (Eds.), 1993. Handbook of Emotions. (New York: Guilford.
  
• MacKinnon, Neil J. 1994. Symbolic Interactionism as Affect Control. Albany: State University of New York Press.
  
• Ortony, Anthony, Gerald Clore, and A. Collins, 1988. The Cognitive Structure of Emotion. New York: Cambridge University Press.
  
• Osgood, Charles H., W. H. May, and M. S. Miron. 1975. Cross-Cultural Universals of Affective Meaning. Urbana: University of Illinois Press.
  
• Robinson, Dawn T., and Lynn Smith-Lovin. 1992. Selective interaction as a strategy for identity maintenance: An affect control model. Social Psychology Quarterly. 55: 12-28.
  
• Robinson, Dawn T., Smith-Lovin, Lynn, and Tsoudis, Olga, 1994. Heinous crime or unfortunate accident? The effects of remorse on responses to mock criminal confessions. Social Forces, 73: 175-190.
  
• Schieffelin, Edward L. 1983. Anger and shame in the tropical forest: On affect as a cultural system in Papua New Guinea. Ethos 11: 181-191.
  
• Shye, S. (Ed.) 1978. Theory Construction and Data Analysis in the Behavioral Sciences: A Volume in Honor of Louis Guttman. San Francisco: Jossey-Bass.
  
• Smith, Herman W., Matsuno, Takanori, and Umino, Michio. 1994. How similar are impression-formation processes among Japanese and Americans? Social Psychology Quarterly, 57: 124-139.
  
• Smith-Lovin, Lynn. 1979. Behavior settings and impressions formed from social scenarios. Social Psychology Quarterly, 42: 31-43.
  
• Smith-Lovin, Lynn. 1987a. Impressions from events. Journal of Mathematical Sociology, 13: 35-70.
  
• Smith-Lovin, Lynn. 1987b. Affect Control Theory: An assessment. Journal of Mathematical Sociology, 13: 171-192.
  
• Smith-Lovin, Lynn. 1994. The sociology of affect and emotion. Pp. 118-148 in Sociological Perspectives on Social Psychology, edited by K. Cook, G. Fine, and J. House. New York: Allyn and Bacon.
  
• Smith-Lovin, L., and Douglas, W. 1992. An affect control analysis of two religious subcultures. In V. Gecas and D. Franks (Eds.), Social Perspective in Emotions, Vol. 1, JAI Press, pp. 217-248.
  
• Smith-Lovin, Lynn, and Heise, D. R. 1988. Analyzing Social Interaction: Advances in Affect Control Theory. New York: Gordon and Breach Science Publishers. (Reprint of a special issue of the Journal of Mathematical Sociology, Vol. 13.)
  
• Smith-Lovin, Lynn, and Tsoudis, Olga. 1993. Empathy in criminal court cases: The effect of emotion displays by the perpetrator and victim ­ An affect control model. Presented at the annual meetings of the American Sociological Association, Miami.
  
• Stein, Nancy L., Tom Trabasso, and Maria Liwag, 1992. The representation and organization of emotional experience: Unfolding the emotion episode. Pp. 279-300 in Handbook of Emotions, edited by Michael Lewis and Jeannette M. Haviland. New York: Guilford Press.
  
• Thoits, P. A. 1989. The sociology of emotions. Annual Review of Sociology 15: 317-342.
  
• Thomas, Lisa, and D. Heise, 1995. Mining error variance and hitting pay dirt: Discovering systematic variation in social sentiments." The Sociological Quarterly, 36: 425-439.
  
• Wiggins, Beverly, and Heise, D. R. 1987. Expectations, intentions, and behavior: Some tests of affect control theory. Journal of Mathematical Sociology, 13: 153-169.