Using Interact


Back Home Up

 

David Heise, 2002 *

CONTENTS

Why Interact

Affect Control Theory

Situations; Sentiments; Equations

Interact

A Basic Analysis

Program-Specific Predictions

Prediction Errors

Analyzing Happenings

Inputting an Interaction; Predicted Behaviors; Predicted Emotions; Social Psychological Considerations

Analyzing Fiction

Analyzing Others

Technical Matters

EPA Profiles; Words from Profiles; Searching; Graphing a Situation; Event Equations; Amalgamation Equations

Sample Projects

End of Romance; A Dealer; The New Look

Bibliographic information about this document

Why Interact?

Some sciences are pretty cut and dried about ordinary things, and the research deals with exotic matters, or with how to rethink everything so simpler laws explain more. When you learn those sciences, you go to laboratories to do experiments that scientists know by heart, experiments that prove how the well-known scientific laws explain things you can see, hear, smell, or feel. You try to make something happen as predicted, and when it happens for you the same as for everyone else, the predictive theory seems convincing. (When you don't get the usual results, you get a poor grade in the course for not mastering the materials!)

Social psychology isn't exactly like that. You learn social psychology through theories that explain some aspect of social relations, but every social psychological theory you learn is both evolving and in dispute. There usually are rival theories that claim to explain the same things, and you end up having to assemble your own framework for interpreting social life by making use of theoretical ideas that are useful to you.

Laboratory exercises aren't much help because most famous social psychology experiments are too complicated and too costly for you to do. Even if you get a Ph.D. in this subject and learn the classic experiments by heart, you don't do most of them! Being without laboratory experience creates a bind. You have to organize and weigh competing theories, but how can you see what's good in a theory unless you try it out yourself? Does the study of social life - of all things - turn out to be purely abstract, even though you're intimately involved with it every day of your life?

There are ways other than laboratory experiments to relate social psychology to your personal experience. You can do some simple field studies, and maybe your observations will favor one theory over another. You can figure out what a theory implies for everyday affairs, and assess whether that prediction fits with what you expect in your own experience. Of course, even a simple social theory gets complex when you try to apply it to several people at once going through a sequence of events. So much goes on in a social interaction that just keeping track of information and doing simple theoretical calculations gets to be a burden. Moreover, some social psychology theories are not all that simple, especially if they try to be specific in predictions. Figuring out the implications in specific situations may get to be plain impossible.

Computers can be programmed to do the work - the data processing, the computations, the logical inferencing - the things that computers do well and fast. With a computer program doing these things, you can define starting conditions and get the computer to pump out the implications of the theory. Then you can compare the outcomes from the computer with your own experience or with new observations in the field, without getting bogged down in details.

That's computer simulation. Surprisingly, a computer simulation is nothing totally new and alien, but simply an aid in applying a theory so you can evaluate it. You know how an electronic calculator lets you do arithmetic faster and more accurately and tackle problems you wouldn't try without it. A computer simulation program works the same way: it lets you examine a theory fast, accurately, and in situations you couldn't handle otherwise. Computer simulations are an extension of the common method of informally evaluating theories by seeing how well they accord with experience.

Some side benefits of computer simulations in social psychology are worth noting. For one thing, a theory has to be well defined in order to make a program out of it, so merely creating a simulation program improves the quality of social psychological theories. That happens to be important in contemporary social psychology, though it wouldn't be if all theories were as explicit as they should be.

The persuasive power of simulations sometimes approaches that of an experiment. That's a plus in a science like social psychology where the educational benefits of laboratory experiments generally are not available to students. After all, any theory is pointless without some means for convincing you that it works.

A simulation program is a working model of a theory's ideas, and it can be used to extend typical scientific procedure in developing hypotheses for test. What happens if something is added or if you take something out? You can do that in the model and see if outcomes still seem okay. Where does the theory give insights that go beyond common sense? You can explore all kinds of circumstances with the model until you find exotic predictions. Once you believe a theory enough so you are not willing to reject it because of strange predictions, then you can use simulations to find strange predictions and see whether they really are true.

One major problem in simulations should be noted. Computers can do so much work at great speed that a simulation can be made to have credibility just by attending to a multitude of details. We might be convinced that a theory is right or wrong more by how well the simulation program handles trivial details than by the underlying sense of what the program is doing. Eliza is a famous program that neatly mimics conversational responses while being utterly mindless about what is being said - Eliza looks like an intelligent conversationalist for a while, yet no theory of human discourse at all is involved in it. So it's a good idea to know the theory that is used in a simulation, and keep paying attention to what it contributes to results.

Alas, social psychological theories rarely are developed enough scientifically to permit computer simulations of social interaction. So from here on we consider just one theory which is so developed.

Interact is a computer simulation program corresponding to Affect Control Theory (ACT). The program lets you set up social interactions by identifying the kinds of people who are present and then reports what kinds of behaviors might occur, what emotions people might feel, how the people might change their views of each other as a result of actions, and how changes in scene might affect behaviors.

Interact , and the theory that makes it go, lets you understand processes that usually are studied separately in social psychology, and the level of prediction is far more explicit than most other theories provide. Millions of social situations can be examined with Interact , far more than you can ever do, so you can run simulations that are of special interest to you, relevant to your own experiences. That allows you to evaluate the theory by comparing its predictions with your own store of knowledge concerning social relations. Once you have some commitment to the theory, you also can use the program to explore meanings of episodes that have you perplexed. Another interesting task is to find the boundaries of where the theory falters, ask why, and think of ways to fill the holes.

Affect Control Theory

Here's the essence of ACT, pure and simple. People try to conduct themselves so their feelings are appropriate to the situation, and if their actions aren't working to do this then they change their views of the situation.

Emotions signal how well events are maintaining our definitions of situations. No emotion is felt when interaction maintains a situational identity in character but to only middling degree - e.g., when a Lover feels fairly good, fairly potent, fairly lively. Emotions are felt either when interactions violate the character of an identity (e.g., a Lover ends up feeling bad, weak, and lively - "panic-stricken") or when interactions confirm an identity both in character and degree (e.g., the Lover feels extremely good, extremely potent, extremely alive - "happy, passionate").

While the capacity for emotional experience is the same for everyone, the way affect attaches to specific objects and behaviors varies across people, and different people may be led to different behaviors, emotions, and interpretations because their affective correlations differ. At the same time, those who share affective linkages share many expectations and judgments about social relations without even having to discuss the matters.

What instincts do for animals, affect does for humans, guiding conduct in directions that are familiar and valuable within a social group. Of course, affective associations are totally learned so human "instincts" adapt to changing conditions, and different groups with their different values provide distinctive patterns of conduct. Moreover, affect governs action within the context of environmental constraint and individual intelligence, so cognition and reason always are a factor in behavior.

Situations

What happens when you go in a place? For one thing, you have to figure out who you are. You may not think about it much. Usually you define the situation and your place in it fast and unconsciously. But you can see what's happening when things mess up. Have you ever walked into a room expecting one group of people - like coworkers - and found someone else instead - like your sweetheart? When it happens you can feel yourself dropping the readiness for some actions and preparing yourself to act in other ways. You're changing who you are in the sense of changing from one role to another, from one social identity to another.

Obviously, another thing you have to do is figure out who the other people are. That may be simply a matter of recognizing people in uniform like a busdriver, or those who always have the same role with you like your car mechanic, or it may be more complicated like figuring out whether another person is being sweetheart or coworker right now when he or she can be both at different times. Who you are depends on who others are, and what roles others take depends on the role you have, so you have to figure these things out simultaneously.

The solution to the puzzle of defining everyone may require more information, like knowing where you are. You and a coworker aren't supposed to act like sweethearts at the place you work; and it's strange to act like coworkers when you and your sweetheart are alone in a cozy romantic restaurant.

You've defined the situation when you can name the setting and the social identities that everyone has. Ordinarily you don't say the names out loud, but you could if someone asked you to describe where you are and who you are and who are the people with you. Interact does ask! That's the way you define a situation in a simulation.

Sentiments

Each identity that you or others take gives a certain feeling because you have a certain attitude about it. Doctor, for example: unless you have a different attitude than most Americans, you think doctors generally are good and helpful, deep and powerful, quiet and meditative. That's your sentiment about doctors, the way you feel in general about them even though you might have different feelings in some circumstances. The general sentiment about Children is quite different: children (for most Americans) are good, but they're small and weak, and noisy and lively. Gangsters provoke still a different sentiment: bad, powerful, and active.

Sentiments have three different components or aspects: we can feel that something is good or bad, that it is powerful or powerless, and that it is lively or quiet. Each of these aspects is a matter of degree, can be greater or less. For example, some things are slightly good, others are quite good, still others are extremely good.

One way of picturing this is to imagine that sentiments are floating around the room you're in. Those things that are very good are up near the ceiling, those that are very bad are near the floor. Things that are powerful are near the wall in front of you, weak things are near the wall behind you. Lively things are on your right, and quiet things are on the left side. Things that don't strike you as either good or bad, powerful or powerless, lively or quiet hang right around the center of the room. So to "see" a Doctor you glance upward to your left; to see a Gangster you look down to your right; and to see a Child you turn your head and look up over your right shoulder.

You have sentiments about ways of acting, too. Look up in front of you to your right, and there's Cheering someone on. Now drop your eyes to the floor along that same corner of the room, and you see Socking someone. Look down behind you on the left; there's Ignoring someone. Look up, forward to your left to see Soothing someone. Each behavior has a sentiment attached to it that reflects how good it is, how powerful, how lively.

The short names for the three aspects of sentiments are Evaluation, Potency, and Activity, and sometimes these are abbreviated further with the initials EPA. We represent a sentiment precisely by measuring it on the three aspects. The custom is to measure everything from the center of the room and use plus units to measure up (goodness), forward (powerfulness), and right (liveliness); minus units are for things that are bad, powerless, or quiet. An EPA profile is a list of three such measures: the first number represents Evaluation, the second is Potency, and the third is Activity.

Equations

Affect control theory works with mathematical equations to predict how events transform feelings toward or away from the sentiments which are evoked in a situation. Equations defining how events change feelings have been developed directly from research on people's responses, and additional equations have been derived mathematically from the change equations in order to define how new events can be formulated to move feelings closer to situational sentiments. But you don't really need to know those details in order to use Interact. Interact doesn't require that you deal with equations at all.

In fact you have enough theory now to comprehend how to use the program. It's time to learn how to make a computer do social psychological analysis!

Interact

Program Interact predicts what events might occur if people have particular identities, which emotions might arise during social interaction, and how people might reinterpret each other as a consequence of events. When you begin Interact the program loads up on human psychology and culture by reading files of equations and words with EPA ratings.

Interact can analyze social interaction among two, three, or four people. Almost always you want to focus on two because a multi-person interaction is so very complex.

Interact lets you define each interactant as a male or a female. The gender you assign someone has several consequences. The most important relates to EPA profiles which are used in analyses. If you say someone is male, then Interact will use EPA data obtained from males in order to represent how the person feels about things. If you identify a person as female on this form, then Interact will use data from females to represent the person's attitudes. Additionally, your choice of gender determines whether a male or a female face is used to display emotional expressions.

Interact alters behavior predictions to take account of past events and to allow for the different people that an actor encounters. Thus you can generate special sequences of events by starting off an incident with some action. You can create other sequences by steering an actor from one interaction partner to another.

A Basic Analysis

Suppose, for example, that we want to examine potential behaviors of a woman named Mary with a man named John. The situation is set up with Mary seeing herself as a woman and seeing John as a man, and with John seeing self as man and Mary as woman. We indicate on the behavior form that we want Interact to figure out optimal behaviors.

Interact's predictions from John's perspective - based on his male sentiments and his personal definition of the situation - define behaviors that John should think are appropriate for himself and behaviors he thinks would be appropriate for Mary. The words in each list are ordered according to how well they correspond to theoretical predictions. Similarly Interact's predictions from Mary's perspective based on her female sentiments and her definition of the situation predict how she thinks John should act and what she thinks she herself should do.

The predictions developed from John's perspective do not agree exactly with the predictions developed from Mary's perspective because of gender differences in sentiments toward "man" and "woman" and gender differences in sentiments toward various behaviors. Gender differences in sentiments sometimes produce more dramatic differences than in this case, and discrepancies in behavioral expectations generally are greater if the people do not agree in their definitions of the situation.

Now, what about the actual predictions - John's acts toward Mary, for example? Interact reports that John might entertain Mary, or hail, amaze, dazzle, acclaim, astonish, exalt, speak to her. Is this what a person identifying as "man" might do to a person he identifies as "woman"? Pretty close, though perhaps a bit exuberant. How about Mary? Interact reports that Mary might speak to John or entertain, compliment, greet, admire, amuse, welcome, amaze him. This is fairly plausible behavior for a woman to a man.

Program-specific predictions

The results reported here are from the version of Interact available in the 1980s. If you run this simulation with the latest version of Interact available to you, you will get results that are similar, but not identical. Why doesn't every version of Interact give the same results?
Equations might be different. For example, early versions of Interact used equations derived with maximum likelihood estimation, whereas later versions often employ equations derived with more stable least-squares estimations.
Sentiment measures might be different. For example, two different dictionaries of U.S.A. sentiments were obtained in the 1970s, and some publications in that era report results from the earlier dictionary rather than the later one.
Cognitive filters for reducing bizarre predictions might be different. For example, early versions of Interact coded relationships as verbal, physical, primary, exchange, managing, fixing, or training, whereas more recent versions use a system based on social institutions: Lay, Business, Law, Politics, Academe, Medicine, Religion, or Family.
Program design might be different. For example, Interact reports words whose EPA profiles are closest to an ideal profile, stopping at an arbitrary criterion, and the criterion has changed in different versions of Interact.
A programming bug might be present in one version. If you believe you have discovered a bug in a current version of Interact, send the details to heise@indiana.edu so that the bug can be corrected.

Results sometimes are less believable than this. When that happens it raises the question: Why would Interact report things that are not true?

Prediction Errors

It is crucial to keep in mind that Interact results are predictions from a theory of human relationships, not reports on what has been observed in actual interactions. Interact results certainly can be wrong for any of the following reasons.
The theory itself - the supposition that people conduct themselves so as to experience events which confirm sentiments about situational identities and actions. If that theory is wrong, then the results from Interact would be wrong, even massively wrong.
Cultural variations in ratings. Interact's predictions derive from EPA ratings of identities, behaviors, modifiers, and settings. The EPA ratings are the way culture gets into the program, and if the ratings are different than those that you would provide, then Interact will be describing a culture which is foreign to you. For example, a set of ratings from Belfast, Northern Ireland, produces some predictions that seem strange to Americans though they are plausible on the whole to an Irishman.
Errors in ratings. Suppose that by a fluke of chance all of the male raters judged "man" as too active, relative to the sense of a man's activity that prevails in the general population. Then Interact, working with the faulty measurement, would make an error in reproducing culture - for example, making predicted behaviors for a man too exuberant.
Errors in equations. Interact incorporates human psychology through complicated equations that describe how feelings about things change as a result of events. These equations have to be defined through research in order to be realistic, but research operations are subject to various kinds of problems that could produce subtle errors in the equations, errors that could cause predictions to be erroneous. Perhaps some alternative equations are better and will become the standard equations eventually. If so, then the equations now in use are producing errors in predictions.
Lexical errors. Interact has to work with words in order to make concrete verbal predictions, and some errors arise because rules governing word usage are not fully understood, thus they cannot be incorporated into the program. For example, you could come upon an Interact prediction that one person "buries" another which is bizarre because "bury" should not be used as a verb describing social interaction. Interact screens words in terms of the kinds of social institutions that are operative, eliminating the worst errors of this kind. However, misusages still creep in and make some Interact predictions look strange.
Misconceptions. For example, Interact predicts that the victim of a deviant act might be given a stigmatized identity by others. You might believe that this is an error because it is unjust - the victim should not be blamed. But in this case YOU would be wrong. Interact correctly predicts derogation of the victim, a phenomenon that actually occurs among humans.

One function of a good theory is to offer new insights and to correct fallacies. So keep an open mind when Interact predicts something about human relationships with which you don't agree. The peculiar prediction may be an idea worth checking out!

Analyzing Happenings

Next we'll see how this allows you to analyze incidents that actually occurred in your life. Then we'll consider a famous novel in order to show how Interact can invent plots for stories.

Incidents which you personally have experienced or observed can be analyzed with Interact. Translating to the framework of the program requires effort and imagination, but the payoff can be an expanded awareness of what happened, clarification of others' conduct, and insights into possible emotions of participants.

Write out a narrative of the episode before you begin in order to solidify your impressions of acts and emotions. Then the task is to find an Interact sequence which fits the incident as closely as possible. In Interact you characterize the people in terms of the identities they had in the situation, and you portray each occurrence as a simple event. You may have to adjust some of your interpretations in order to obtain congruence with the Interact reconstruction, but often as not you'll find this enlightening rather than thwarting.

Here is a narrative of something that actually happened. The quality of the data is typical of what you might have in analyzing personal incidents - a narrative constructed more than a day after the incident. That is better than nothing even though it is not very good data by research standards - it would be better to work with notes made minutes after the happening, better still to work with notes made by an observer during the happening, and better still to work with a transcript made from a sound-image recording.

The professor designed a new course with substantial reading assignments, field studies, learning how to use microcomputers. Today the first paper was due, 1500 words concerning a personal experience, written with a word processing program and incorporating auxiliary computer analyses. However, the professor bungled, forgetting to provide adequate guidance on how to get papers printed.

A student enters the class late, creating some disturbance, and explains brusquely that he had trouble getting his paper printed. He sits down, and as the professor tries to get the discussion back on track after the student's entrance, the student asks, "Just how much time were we supposed to put into this paper?" The professor stares at the student and quietly, very quietly, says, "As much as it takes." The student mutters, "Oh. All right." He looks away and makes no further disturbance.

Can something like this be analyzed with Interact? Let's try.

Inputting an Interaction

Who are the characters in this episode? A psychologist specializing in personality might want to administer a personality inventory and a few projective tests in order to comprehend their traits and motives before saying anything about the episode. But the theory behind Interact claims that situational roles determine people's conduct, and all we need to know about these two people is what social identities they had in the circumstances. Inferring people's identities sometimes can be a problem, but not here since the whole episode took place in a university classroom. It is reasonable to suppose that the student was being a "student" and the professor was being a "professor". Thus on the form for identifying self and others, we select student as the identity of one person and professor as the identity of the second person. We also make sure that both characters are male so that they have the correct sentiments guiding them.

Next we have to incorporate the circumstances given in the first paragraph of the narrative of the incident in order to set the scene. We incorporate the background by forcing events to happen.

The demanding professor with his assignments and out-of-class requirements might be said to have "worked" his students. However, that probably is not the way the student saw it; he evidently believed the professor "overworked" his students - a less desirable act. We can represent the student's perception on the form for defining events by making the student's first event "professor overworks student". Additionally we can represent the professor's viewpoint by defining "professor works student" as the professor's first event.

The professor's failure to provide guidance about how to print papers might be described as letting the students down. However, Interact doesn't know the behavior "let down", and we have to find a substitute. The word "forget" is used in the description, and the Interact list of behaviors does contain that word, so we might say that the professor's second event was that he forgot his students. However, "forget" doesn't seem assertive enough to reflect the student's perception of what happened. Let's suppose the student's second event is the professor stymies the student, which is possible because the list of behaviors in Interact contains "stymie".

Predicted Behaviors

Now that we've set the scene, the key question is whether Interact will predict what the student and the professor did.

On the form for implementing events we select the student's viewpoint and implement the first event - professor overworks student. That allows us to see how the student might have responded to that event alone. Surprisingly, Interact predicts that the student still would act positively toward the professor. The behavior predictions are debate with, titillate, dazzle, astonish, astound, kid, flatter, question. Shouldn't an overworked student be more aggressive? In fact, only the most subtle aggressiveness is all that is possible. The student cannot bluntly act out his negative feelings because he has to maintain positive identities - his own identity of student and the other's identity of professor - and hostile acts would create negative impressions of both.

Implementing "professor stymies student" leads to Interact's predictions of what the student might do after both events.

The predictions are that the student might challenge the professor (or titillate, fox, debate with, dazzle, astonish, influence, rally him). These acts are not really hostile, but some are distinctly assertive. Moreover, "challenge" is an adequate description of the student's question about time investment for the paper.

Let's select that behavior, so "student challenges professor" is the next event in the sequence, from the student's viewpoint.

The next round of events is supposed to predict the professor's response. The predicted behaviors are: soothe, consider, calm, console, caution, bless, counsel, explain (to). Once again we see surprisingly positive behaviors because the professor has to maintain the goodness of his role and the student's. The word "bless" seems out of place for lexical reasons: it is a religious act that should not arise in secular relationships.

Anyway, the word "caution" is there, and that is a reasonable designation for the professor's slow and quiet reply to the student. The professor was cautioning the student to not continue the challenge.

This translation of the professor-student incident into the framework of Interact seems to have been straightforward with few hitches and impediments. Is analyzing interaction with Interact always so easy? No, nor was it really easy this time. In order to keep my exposition on track, I omitted mentioning some of the complications.

Finding a word for the student's view of the professor's second background behavior was problematic. The designation of "stymie" resulted after analyses were run with two other possibilities: hassle and discourage.

Being overworked and "hassled" led to these predicted behaviors for the student: debate with, titillate, tantalize, astonish, astound, fox, question, influence. I could have used "question" to represent the student's opening sally, but that word does not capture the aspect of challenge that was evident in the student's interruptions. Thus "hassle" was discarded because it veered Interact away from what was observed, notwithstanding that the word might capture important elements of how the student viewed things.

Being overworked and "discouraged" by the professor led to these predicted behaviors: challenge, titillate, fox, debate with, influence, dazzle, astonish, astound. There is the desired "challenge" and the rest of the analysis works out, too. The trouble is that "discouraging students" doesn't accurately describe what the professor did.

I needed a word like "discourage" in terms of affective impact but one that described more accurately the way the student viewed the professor's lack of guidance. Using procedures to be described soon, I settled on "stymie" as the best of the choices available in Interact (though still not ideal).

The word "caution" shows up among predictions for the professor as required for this interaction, but other acts rank above it - like soothe, calm, console. Why didn't the professor enact one of these? (In fact, at the opening of the class the professor did engage in such acts with other students who arrived on time.) Why did the professor take a more active stance after the student challenge? I tried a number of variations in definitions of prior events, but never could get "caution" to be the highest ranking predicted act from the professor's perspective; and sometimes I lost the prediction of cautioning all together. My best guess is that this slight deviation from reality - the high ranking of soothe, calm, console after the student challenge - perhaps arises from errors in measuring the activity levels of identities and behaviors, or from errors in the equations representing how activity is processed psychologically.

Predicted Emotions

Now that we have a sequence of behaviors for Interact which corresponds reasonably well with the observed incident, we can go back and examine the predicted emotions.

Interact predicts that working students is basically an emotionally-neutral act for the professor. Nor does the professor suppose that students will be upset: possibly awe-struck, maybe anxious, but largely emotionally neutral.

But the student experienced a different event - one of overwork - and his predicted emotions are quite different: awe-struck, overwhelmed, self-conscious, fearful, apprehensive, lovesick, nervous, impatient. Moreover, Interact predicts that he thinks the professor - in overworking his students - must be feeling contemptuous, scornful, unfriendly, spiteful. After the next event Interact predicts that, since the professor both overworked and stymied him, the student should feel afraid, overwhelmed, insecure, flustered, embarrassed, terrified, afraid, self-pitying; and he should see the professor as being contemptuous, vengeful, spiteful, hostile, outraged, grouchy.

Returning to the professor's viewpoint, the professor who worked his students and then forgot them is predicted to feel unfriendly, disturbed, irritable, contemptuous, bitter, cynical, grouchy, irritated. Now it is time to confess that I myself am the professor in this incident because I want to offer judgment on the accuracy of the predicted emotions for the professor. They are close - especially contemptuous, bitter, cynical, grouchy, irritated - but with a twist. I saw myself victimized by university personnel rather than generating the emotions myself. "How am I supposed to teach an advanced technology course when this university doesn't give me facilities I need? Nobody provides me with information on how students can print their papers so naturally I couldn't tell the students how to do it. All the university's fault! Bah!" Of course, this was a rationalization. Sometimes I was painfully aware that I'd forgotten the students' predicament, but every time I thought of my lapse I remembered a different incident in which I could blame someone other than myself for the negative emotions I felt.

Right after the challenge the professor's emotions are predicted to be: horrified, frightened, scared, terrified, jealous, afraid, frustrated, flustered. I recall flashes of feeling frustrated and flustered and tinglings of stronger fears. "Horrified" is appropriate in the instant when I thought, "My god, we're going to have a scene!" The student is predicted to feel apprehensive, shocked, self-conscious, displeased, irked, resentful, fed up, annoyed; and several of those emotions seem plausible.

The professor's emotions as he engages in his cautioning are predicted as: self-conscious, apprehensive, fearful, shocked, overwhelmed, uneasy, lovesick, heavy hearted; and in fact self-conscious, apprehensive, and uneasy are right on the mark in describing how I felt as the incident ended. The student is predicted to feel shocked, uneasy, heavy hearted, fearful, discontented, regretful, self-conscious, displeased, and at least some of these feelings seem likely.

Social Psychological Considerations

The key to analyzing this incident with Interact was understanding that professor and student had different interpretations of key events. An act commonly can be evaluated in different ways - like working someone versus overworking someone, and the agent and the recipient of the act arrive at different evaluations because of their different burdens in the event. Then the differing interpretations strain social interaction and undermine mutual confirmation of identities. Even slightly negative interpretations of events engender negative emotions, and after that the distressed person's efforts to regain equilibrium become disturbances for the other.

Interact analyses sometimes suggest an abundance of negative feelings accompanied by barely discernable deviations from positive action. This is non-intuitive, yet supported by empirical studies. Intensive analysis of ordinary social interaction reveals a startling amount of aggression and negative feelings, masked superficially but evident in slow motion viewing of image recordings. The phenomenon is so pronounced that it actually creates ethical problems in reporting results because it makes people look bad!

Analyzing Fiction

Certain themes appear over and over in fiction and folktales. Such stories may have an affective basis, with the stories exploring the experiences and emotions one has in deviant relationships. Tales of vampires offer a good arena for illustrating this because we have "vampire" in the Interact dictionary, and because a standard rendition of the tale is available in Bram Stoker's Dracula.

In Stoker's novel, a young man named Harker is sent from London to a wild area of central Europe to engage in business with Count Dracula. The opening pages of the novel develop Harker as a pleasant "person of no importance" - a solicitor's clerk. We will use the identity normal-man in simulations. Dracula's identity as an evil and powerful being is evident almost from the start, and we soon learn specifically that he is a vampire.

The first encounter between Dracula and Harker occurs at night on a carriage ride to Dracula's castle. Dracula drives back and forth over the same road, making sudden turns, and stopping and jumping from the carriage in order to chase blue flames. Wolves circle the carriage and Dracula disbands them with a command. At the castle he leaves Harker waiting alone in a gloomy courtyard for an extended time. Harker refers to his emotions in the following terms: frightened, a sick feeling of suspense, afraid, crowded by doubts and fears.

Before we consider more of the story get Interact running, and let's try to simulate this. The analysis is a good place to try 3-person interactions because Harker will encounter female vampires, too, and his interaction with them is primed by his previous experience with Dracula. Make the first two interactants males and the third female. Name the first "Harker" and make the identity "man" modified by "normal". Name the second actor "Dracula" and give him the identity of "vampire" with no modifier. Name the third actor "Ladyvamp", and make her a "vampire". We will assume that all of the actors see each other the way each sees the self.

From each interactant's perspective, set up first events with unspecified behaviors so that Interact will find behaviors that best maintain the identities in each dyad. Here are the results from Harker's perspective:

Harker might pacify Dracula (or appease him). ___
Harker might query Ladyvamp (or order her). ___
Dracula should spurn Harker (or unnerve him). ___
Dracula should judge Ladyvamp (or detain her). ___
Ladyvamp should idolize Harker (or look_at him). ___
Ladyvamp should humble Dracula (or admonish him). ___

[In the 1980's version of Interact] we get only two predicted behaviors for each relationship because of space problems. However, I also ran the analysis with just two people at a time and results from those analyses expand the information in the three-person analysis. Here are the predictions if we run only Harker and Dracula together.

Harker might pacify Dracula (or appease, query, entreat, hush, order, beseech, psychoanalyze him). ___
Dracula should spurn Harker (or unnerve, repulse, confuse, disregard, vex, hoodwink, hinder him). ___

Some of the predicted behaviors certainly seem to characterize Dracula's initial actions. Moreover, Chapter 2 of the novel describes Harker's further observations and conversations with Dracula - a variety of behaviors that are reasonably summarized by some of the Interact predictions for the first act of a normal-man toward a vampire as listed above.

Returning to the three-person analysis we implement the first event as Dracula "unnerve" Harker. Dracula unnerving Harker leaves Harker with the following predicted emotions: awe_struck, overwhelmed, selfconscious, apprehensive, melancholy, shocked.

Chapter 2 ends with Harker exploring the castle and discovering that "The castle is a veritable prison, and I am a prisoner!" This corresponds approximately to the next predicted acts of a vampire toward a normal-man: bind, obstruct, and confine.

In Chapter 3 Harker sleeps in an unused part of the castle in opposition to the count's explicit warning about doing so. This corresponds to the Interact prediction that Harker would rebuff the vampire at this point.

Harker wakes to find himself in the presence of three beautiful female-vampires. "They came close to me, and looked at me for some time, and then whispered together." One of the girls said, "He is young and strong; there are kisses for us all." Harker watches through his eyelashes as "The fair girl advanced and bent over me .... Lower and lower went her head as the lips went below the range of my mouth .... I could hear the churning sound of her tongue as it licked her teeth and lips .... I closed my eyes in languorous ecstasy and waited - waited with beating heart."

Interact predicts the following acts for a female-vampire who encounters Harker at this point: applaud, look at. Now implement Ladyvamp "look_at" Harker, and then one of the next predictions: Ladyvamp "initiate" Harker. We find Harker feeling cowardly. This is not the strange ecstasy described in the novel - the simulation has Harker feeling weak and negative rather than weak and positive. But then English doesn't provide any word for a good weak emotion, so Interact couldn't present one. The female vampire is predicted to feel amused, glad, relieved, cheered, delighted, charmed, passionate, touched.

At that instant Dracula appears on the scene. He angrily forces the girls away from Harker saying, "How dare you touch him, any of you?" He promises the girls they can have Harker after Dracula is done with him. The girls laugh and accuse Dracula of never loving, and one says, "Are we to have nothing tonight?" Dracula allows them to take away a child whom he brought in a bag.

According to Interact, the vampire's predicted actions toward the female-vampires at this point are to reproach and discourage. Their predicted behaviors toward him are to order and patronize.

The novel continues to explore the relationships of a vampire with more people and the relationships of his victims with one another. Dealing with the full story is more than we want to do and beyond the capabilities of Interact in its present form. We've seen enough: that a vampire tale is the story of what people experience while confirming the identity of a character who is extremely bad, powerful, and dead.

Analyzing Others

Interact can be used to help you comprehend motives and feelings of people who engage in behavior which is mysterious to you. The procedure is this.

Identify someone who engages in interpersonal activities that are deviant or odd, and through observations and/or conversations with that person make yourself informed about the activities and about how the person views him or herself when engaging in those activities. Since you can't assume that you know much about it, part of the task is distinguishing details of the activities and the roles. However, you especially want cues about how your consultant judges the acts and identities. Additionally, get a thorough narrative of one typical experience: actors' identities, what interpersonal actions occurred, and how your consultant felt as each event unfolded. Translate the narrative into simple events which you can analyze with Interact, each event consisting of actor, behavior, and recipient.

Use the Java Rater or the Javascript Rater to rate the deviant identities and behaviors from the viewpoint of the other. You cannot assume that the consultant has the same attitudes as you do, so do this carefully: enter ratings you can justify by reference to something your consultant said or did. Go through the ratings several times, preferably at different times, in order to improve your measurements by averaging.

Use the option in Interact for importing data, and add your ratings to the dictionaries. You can use the new words in Interact analyses until you quit the program.

Set up an analysis with the new identities and reproduce the narrative, forcing whatever behaviors were involved, including the special deviant acts. The goal here is to check the emotions which your consultant reported during the incident as a test of your ratings. If you do not get the right emotions you should treat your ratings as erroneous and try to revise them so that they work reasonably well.

When you can simulate the incident, you are ready to explore your consultant's mind in rich detail. Here are some things to examine. What emotions did he or she think other people were feeling? What kinds of behavior might be expected in the incident aside from those in the narrative? How would the consultant reidentify someone who engaged in the crucial deviant act? How might the consultant in his or her deviant role behave with others who weren't in the narrative?

You may want to return to your consultant to check what he or she says about these things. It also may be interesting to examine the same things using the usual profiles in Interact in order to see the responses of normal people.

EPA profiles for some identities and behaviors as rated by people in special sub-cultures are available at this site, ready to import into Interact.

Technical Matters

Interact is designed so that you don't have to worry about technical matters very much: you describe people and events in words, and the program makes all of its predictions in words, too. But everything that Interact predicts about people and social life is figured out at a completely different level by translating a situation into measurements of sentiments and feelings.

In fact, Interact understands very little about the words you use to describe a scene. The program simply uses the words to render the scene as a numerical problem - a representation that literally could be drawn on a graph. Then Interact calculates with equations that describe how attitudes and sentiments combine and change, and thereby Interact comes up with predictions. The predictions are numbers, but Interact translates numerical predictions back into words.

EPA Profiles

For example, suppose you specify that Mary is a "woman." Interact knows very little about women - nothing about what is required to be a woman or what responsibilities women have. All that Interact knows is how people feel about women - that on the whole, women are quite good, somewhat powerful, and somewhat lively. This information is contained in an EPA profile which summarizes ratings from a number of people, a profile like you can produce yourself using the measurement program. The EPA profile for "woman" is what Interact uses to set up a situation with a woman in it, and that EPA profile is all that Interact uses to make predictions about womanly behavior.

Here are the EPA profiles for "woman" stored inside the computer.

male: 2.34 0.43 1.14; female: 1.74 0.67 0.85

The first set of numbers summarizes ratings from about 25 males, and these numbers mean that on the average the males rated a "woman" as between quite and extremely nice, between neutral and slightly powerful, and slightly lively. The second set of numbers represents the ratings from about 25 females: they rated "woman" almost the same as males did - a little less nice, a bit more powerful, and less lively, but not enough to make a difference. (The corresponding numbers for males and females would have to differ by about .8 or more to make a difference.)

Here are the EPA profiles for "friend."

male: 2.66 1.81 0.92; female: 3.48 1.21 0.25

In words, friends are extremely nice, slightly to quite powerful, and slightly lively. The ratings from males and females are similar, but females think a friend is even nicer than males do, and the female ratings also represent a friend as weaker and quieter than the male ratings do.

Suppose we were to set up a woman-friend interaction. If we specify that both parties are female, then Interact would represent the scene using female profiles: 1.74 0.67 0.85 for woman and 3.48 1.21 0.25 for friend. From those numbers Interact produces predictions about the standard behaviors of a woman to a friend

dazzle, entertain, rally, amuse, amaze, exalt, awe

and about a friend's behaviors to a woman

please, cheer, congratulate, aid, welcome, apologize to, assist.

Interact doesn't arrive at names for behaviors directly. It computes an EPA profile to represent the ideal behavior of a woman to a friend or a friend to a woman, then searches for behaviors with a profile like that. For example, the ideal profile for the behavior of a friend to a woman is 2.3 1.3 0.3 (and you'll learn how to find that out pretty soon). Evidently the behavior of pleasing must have an EPA profile close to 2.3 1.3 0.3 since "please" was the top ranking act of friend to woman.

Let's check that. The EPA profiles for the behavior "please" are as follows.

male: 1.70 1.06 0.25; female: 2.06 1.06 0.21

Sure enough, the female EPA profile for "please" is close to the computed EPA profile for a friend's ideal behavior toward a woman, though not exactly the same.

If you select the option where Interact reports EPA profiles then EPA profiles are reported every time a word in an Interact dictionary is selected and every time a prediction is displayed. For example, with the option on in the Janet-Mary interaction you will see that the EPA profile for woman is 1.73 0.68 0.85, and the profile for friend is 3.48 1.20 0.26. Also you will see that the ideal EPA profile for Janet's action after Mary comforted her is 2.08 1.51 1.61. These numbers are the inputs and outputs of the mathematical computations that give Interact its capacity for intuition.

Words from Profiles

When making predictions Interact searches for words that match a particular profile, and you can use the same routine yourself to get a list of words that match a profile like 2.3 1.3 0.3. Here is what results when you enter the EPA profile on the search form.

male: enjoy, like, understand, satisfy, caress, help, marry, fondle.
female: hug, help, like, aid, smile at, please, trust, enjoy.

The male and female lists are different because males and females rated the behaviors differently.

But wait! Shouldn't the female list be the same as when we had Interact predict a friend's behavior toward a woman? "Please" is supposed to be first. And where did "hug", "help", etc. come from? Here's the explanation. Interact screened all of the other kinds of behaviors out of its list of predicted behaviors except those that logically fit the relationship between a woman and a friend. Suppose we rerun the friend-woman analysis with all filters open? That allows all words to appear, and then Interact's predictions about a friend's behavior to a woman are: hug, like, help, enjoy, marry, please, cheer, congratulate. Well, that's better, but still it is not the same list as we got by direct search. Why? Because the numbers we entered directly were rounded to one decimal place and not as precise as the numbers Interact used. Since Interact searched for words matching a slightly different profile, the ordering of words is different, and Interact found "marry", "cheer", and "congratulate" instead of "aid", "smile at", and "trust." In fact, we, too, should find "marry", "cheer", and "congratulate" if we change the profile just a little. Try increasing the evaluation of the search profile by 0.1. That is conduct a search with the profile 2.1 1.3 0.3. You'll find that "cheer", "congratulate" and "marry" do appear in the female list, so the extra words which Interact found really are close to the profile we entered.

Searching

The search function in Interact is especially useful when you are looking for an identity or a behavior that has a certain feeling.

For example, in the last lesson remember that I wanted to replace the word "discourage." To do this I first found out the male EPA profile for "discourage", -1.53 0.53 -0.28. Then I  entered the profile for "discourage" on the search form, and snooped around until I found the word "stymie" which has a fairly similar profile, -1.14 0.99 0.18.

You can rummage in a similar way among identities. That is, define a profile that has the right feel for your purposes, and then see what identities are available in Interact to approximately represent that feeling.

You can search for modifiers or settings in the same way.

Graphing a Situation

I mentioned that intuitions about a scene could be graphed, and it is worth actually doing that in order to get a better sense of how Interact works. Here's an incident to consider.

Janet's boyfriend argues with her and leaves, so Janet feels sick-at-heart as Mary arrives. Mary, being a friend, comforts Janet. Janet sighs and cracks a joke.

This is a hypothetical happening, but I have set it up to correspond to a laboratory experiment which was designed to test the theory behind Interact. In the experiment, a subject was made to feel down (sick-at-heart,  deflated) by an interaction with a nasty secretary - really an actress hired for the experiment, and then the subject had to interact with another actor who had been introduced as a fellow student. The experiment showed that people who are feeling down about something pull out of their gloom and act pleasant when they are with someone they value - like a fellow student, though they don't do this when they are with someone they don't value - such as the same actor described as a "delinquent."

The accompanying graph represents the situation for Janet and Mary.

Measurements in the vertical direction represent Evaluation, and measurements in the horizontal direction represent Activity; we'll forget about Potency for now, because Evaluation and Activity are the interesting measures in this case.

Janet, in her identity as a "woman", appears in the middle right section of the graph - evaluation of 1.7 and activity of 0.9. Mary, as a "friend", appears at the top of the chart - evaluation of 3.5 and activity 0.3.

Janet after her argument with her boyfriend is "sick-at-heart", and thereby is not nearly so nice and lively as a woman ordinarily is. In fact, if we had people rate "a woman who is sick-at-heart" the evaluation and activity ratings would average out to barely positive on both measures. Thus Janet's transient state is represented on the graph at the bottom middle.

At that point Mary comforts Janet. As you can see on the graph, the act of comforting is rated as extremely good and a little quiet. In fact, comforting is a bit more good and more quiet than the usual acts of a friend to a woman (like "please" or "cheer"), and that reflects Mary's response to Janet's negative condition. Mary has to act especially nice and quiet in order to improve Janet's mood and in order to maintain her own self-esteem while dealing with a gloomy woman.

After Mary's comforting, we get Janet's response to being comforted. The comforting worked a little because Janet has moved up on the graph to a slightly better state, from being a "sick-at-heart woman" to being a "melancholy woman." Janet is still not cheerful like a woman should be, but she is a bit better off than before.

Janet engages in one of her predicted behaviors toward Mary at this point: amuse. As you see on the graph, amusing someone is quite a nice act and lively. In fact, Janet now is trying to act nicer than usual toward her friend because that is a way of pulling herself out of her negative emotion: Janet amuses her friend even as she herself feels melancholy in an effort to restore her own usual cheerfulness. She also is acting nice in order to maintain the goodness of her friend.

The tactic partially works. After amusing Mary, Janet's emotional state has improved again, up to feeling "relieved."

Janet and Mary would have to go through a few more rounds of interaction before any consequences of the emotion produced by the incident between Janet and her boyfriend were gone. Moreover, once she stops interacting with Mary, Janet might recall the incident with her boyfriend and return to feeling sick at heart.

Event Equations

Another option in Interact let's you look at the equations which are used in the computations. For example, these are the first few lines of the U.S.A. male actor-behavior-object (ABO) impression-formation equations.

Z000000000 -0.25 -0.09 0.07 -0.15 0.03 -0.02 -0.09 -0.38 -0.03
Z100000000 0.44 -0.02 0.05 0.11 0.03 -0.01 0.01 0.00 -0.01
Z010000000 0.00 0.59 -0.05 0.03 0.15 -0.07 0.00 -0.06 0.00

Impression formation equations are the heart of Interact. They describe how an event changes feelings about a person, and also how an event changes feelings about behaviors and settings. "ABO equations" deal with events specified in terms of actor, behavior, and object person.

Each column of decimal numbers in the table represents a different equation, and the numbers are the coefficients for different terms in the equation. For example, the first column of decimal numbers defines an equation for predicting how an actor will be evaluated after an event. The second column gives an equation for predicting how powerful an actor will be after an event. In the case of ABO equations, there are columns defining how to predict the EPA outcomes for actors, behaviors, object persons.

The column of zero-one numbers, preceded by "Z" identifies the terms that are in the equations, as follows.

if the first digit in a line has the value 1, then the term of the equation defined by that line involves the pre-event evaluation of the actor, Ae
" 2nd digit " pre-event potency of the actor, Ap
" 3rd digit " pre-event activity of the actor, Aa
" 4th digit " pre-event evaluation of the behavior, Be
" 5th digit " pre-event potency of the behavior, Bp
" 6th digit " pre-event activity of the behavior, Ba
" 7th digit " pre-event evaluation of the object, Oe
" 8th digit " pre-event potency of the object, Op
" 9th digit " pre-event activity of the object, Oa
" none of the digits " equation constant
" more than one digit " multiplication of several pre-event quantities

Now let's put together the equation for predicting the outcome evaluation of an actor as a result of an event, Ae'. The column of zero-one numbers, the term that those numbers define, and then the first column of decimal numbers are as follows.

Z000000000 constant -0.25
Z100000000 Ae 0.44
Z010000000 Ap 0.00
Z001000000 Aa 0.01
Z000100000 Be 0.41
Z000010000 Bp -0.04
Z000001000 Ba -0.10
Z000000100 Oe 0.02
Z000000010 Op -0.02
Z000000001 Oa -0.01
Z100100000 AeBe 0.05
Z100010000 AeBp -0.03
Z100000100 AeOe 0.00
Z100000010 AeOp 0.01
Z010100000 ApBe 0.01
Z010010000 ApBp 0.00
Z010000010 ApOp 0.02
Z010000001 ApOa -0.01
Z001001000 AaBa 0.00
Z000100100 BeOe 0.13
Z000100010 BeOp -0.06
Z000010100 BpOe -0.06
Z000010010 BpOp 0.07
Z000010001 BpOa 0.01
Z000001010 BaOp 0.03
Z100100100 AeBeOe 0.03
Z100010010 AeBpOp 0.02
Z010010010 ApBpOp -0.02
Z010010001 ApBpOa 0.02

We start off building the equation by representing the equation constant like this:

Ae' = -0.25

Ae is the evaluation of the actor before the event; it is associated with a large coefficient, 0.44, so evaluation of the actor before an event has a major impact on evaluation after the event. By including the term for Ae the equation becomes:

Ae' = -0.25 + 0.44Ae

The actor's potency before the event doesn't affect evaluation afterwards - the coefficient for Ap is zero. In fact, let's consider any coefficient with a magnitude of 0.10 or less as too small to count for our heuristic purposes at the moment. Then  the next term that is important is Be with a coefficient of 0.41. The equation becomes

Ae' = -0.25 + 0.44Ae + 0.41Be

Continuing this way down the column we get the following equation for predicting the evaluation of the actor after an event has happened.

Ae' = -0.25 + 0.44Ae + 0.41Be - 0.10Ba + 0.13BeOe

Note that BeOe - the product of behavior evaluation with evaluation of the object person - is positive if someone does something nice to a nice person or does something bad to a bad person; the quantity is negative when evaluations of behavior and object person do not correspond - when one evaluation is negative and the other is positive.

Thus the equation says that good impressions of actors are created when the actors are good to begin with, when they engage in good and quiet behaviors, and when behavior evaluations are appropriate to the evaluation of object persons. Remember in the Janet-Mary analysis: Mary comforted Janet which was an unusually nice and quiet act from a friend. Now you see why Mary did this. Mary had to act nice in order to confirm being a friend - the goodness of behavior has a very strong effect in determining what kind of impression is created. However, being nice to a gloomy woman friend gained nothing for Mary in terms of matching act to object (the BeOe effect), so Mary had to build up extra goodness by choosing an act that is especially good and quiet.

We have achieved what we wanted - a general sense of how Interact arrives at predictions. Of course, Interact computations do not drop terms with coefficients having a magnitude less than 0.10 because many such terms actually are important in representing human psychology. The following full equation with all terms is the one employed in Interact.

Ae' = -0.25 + 0.44Ae + 0.00Ap + 0.01Aa + 0.41Be - 0.04Bp - 0.10Ba + 0.02Oe - 0.02Op - 0.01Oa + 0.05AeBe - 0.03AeBp + 0.00AeOe + 0.01AeOp + 0.01ApBe + 0.00ApBp + 0.02ApOp - 0.01ApOa + 0.00AaBa + 0.13BeOe - 0.06BeOp - 0.06BpOe + 0.07BpOp + 0.01BpOa + 0.03BaOp + 0.03AeBeOe + 0.02AeBpOp - 0.02ApBpOp + 0.02ApBpOa

Amalgamation Equations

Interact also employs impression-formation equations describing how modifiers combine with identities. For example, here is the short-hand representation of the U.S.A. male equations.

Z000000 -0.26 -0.18 0.07
Z100000 0.67 -0.15 0.05
Z010000 -0.29 0.76 -0.09
Z001000 -0.11 0.06 0.67
Z000100 0.47 -0.02 0.01
Z000010 -0.02 0.56 -0.09
Z000001 0.00 0.07 0.67
Z100100 0.12 0.00 0.00

The three columns of decimal numbers define the equations for predicting the evaluation, potency, and activity of a combination. The first three Z digits refer to the evaluation, potency, and activity of the modifier. The second three digits refer to the evaluation, potency, and activity of the identity.

For example, the equation for predicting the evaluation of a modifier-identity combination is:

Ce = -.26 + .67Me - .29Mp - .11Ma + .47Ie - .02Ip + .00Ia + .12MeIe

In words, the evaluation of the combination, Ce, is predictable by summing approximately two-thirds of the modifier evaluation, Me, and half of the identity evaluation, Ie, and then adjusting a little for modifier potency and activity, Mp and Ma, and for evaluative consistency between the modifier and identity, MeIe.

Interact frequently turns the amalgamation equations inside out in order to find modifiers - for example, in order to find emotion terms that can be combined with a person's identity in order to describe the person's transient state.

Sample Projects

End of Romance

A Dealer

The New Look

 

 

 

Note. From David R. Heise and Elsa Lewis, Introduction to Interact, documentation for programs Interact, Tech, and Attitude, distributed by Wm. C. Brown Publishers, Dubuque, Iowa 1988-1993.

 This  revision for the World Wide Web - in contrast to the original text - forgoes:
Step-by-step directions for setting up analyses, since procedures vary from one version of Interact to another.
Topics that are not relevant to the version of Interact available in 2002 (finding an action to produce an emotion; and producing random output in Interact).
The chapter on program Attitude, which is replaced by the program Surveyor documentation for Project Magellan.
The chapter on Roles, which was revised for the presentation on roles in the affect control theory tutorial.
The chapter on Emotions, which was revised for the presentation on emotions in the affect control theory tutorial.
The chapter on Reidentification, which was revised for the presentation on reidentification in the affect control theory tutorial.
The chapter on Deviance, which was revised for the presentation on deviance in the affect control theory tutorial.
The chapter on Sub-Cultures, which was revised for the presentation on sub-cultures in the affect control theory tutorial.
The References section, which is replaced by the list of ACT readings.
 

 

 

Back Home Up

 

URL: www.indiana.edu/~socpsy/ACT/InteractTutorial/using_interact.htm