In this book we have emphasized the importance of group process variables. They are the necessary links between input and output variables. To describe and explain group functioning completely, a scientist must consider process variables. In the early days of small group research, researchers did not note this fact. They would often discuss the importance of group interaction but they rarely studied it. Even today's scientists continue to ignore process variables too frequently. As we have shown in this book, this shortcoming often leads to research results that we cannot interpret unambiguously; findings often are not clear-cut.

As we mentioned in Chapter 1, the best way to evaluate process variables is to perform a content analysis. A content analysis consists of direct studies of group process. Throughout this book we have discussed the results of many content analytic studies of group discussion. In reading about these studies you may have become aware of how important it is to study group process through content analysis. Our goal in this appendix is to present an introductory discussion of how scientists can perform this kind of research.

To perform a content analysis successfully, a researcher must go through a great deal of preliminary preparation. Part of this preparation consists of various series of questions and answers. An researcher needs to ask what to study and how to study it.

The series of questions regarding what to study comes first. The answer to this first question depends on the researcher's aims. For example, researchers may be interested in evaluating a theory about group discussion. As you recall, in Chapter 8 we discussed how Bales did this when he formulated his research strategy. Bales believed that group output was dependent on how well the group managed the "equilibrium problem" between task and maintenance goals. Bales designed a content analytic method for studying group process that allowed him to evaluate this belief.

Unlike Bales, most practitioners have no theoretical axe to grind. Instead, the practitioner merely wants to make an insightful analysis of a group. When this happens, it is a good idea for the practitioner to make a preliminary observation of a group meeting. The intent of this observation is to note anything the group does that is interesting or potentially significant. For example, an observer might see different group problems that include conflict over status and leadership or difficulty in carrying out procedural steps. On the other hand, she or he could note something the group does particularly well. For instance, the group may take overt steps to include every member in its discussions.

In either case, to perform research the practitioner must come up with a list of behavioral indicators of the feature he or she wants to study. These are activities the group members perform that clearly display that feature. Behavioral indicators may either be verbal or nonverbal. For example, Bales believed that utterances that "showed solidarity," "showed tension release," and "showed agreement" were behavioral indicators of positive group maintenance.

Once a researcher has decided what to study, he or she then moves on to the next series of questions. This series concerns how to perform the research. In essence, the person moves from the "what-to-study" question to the "how-to-study" question. The first part of answering the how-to-study question involves the construction of an appropriate coding scheme. A scheme is a classification system for the behaviors under study. The coding systems can work as guides for the research. In Chapter 8 we discussed Bales's coding scheme for studying the behavioral indicators of positive and negative maintenance and task work.

Coding schemes can range from the very simple to the very complex. For example, suppose the researcher decides to study the "sound-silence" patterns of the members of a group. Most simply, these patterns consist of the times when each member is talking and the times when each is silent. For our example of sound-silence patterns, the basic coding scheme is self-evident. Either a member is talking, or he or she is silent. It is true that there are some complications to this scheme. For instance, there are "back-channel utterances," such as the "uh-huh" that one member might say while the other is talking. However, we need not address these complications here. Incidentally, "sound-silence" patterns may appear trite, but they are actually quite significant. It is valuable to continue to research them. For example, as we discussed earlier, talkative group members tend to be more influential than nontalkative participants. The group also sees talkative people as "leaders" more often than quiet members.

In contrast, Bales's coding system is relatively complex. As you recall, it consists of four counterbalanced groupings, each of which had three categories. A study of Bales's organization should reveal how difficult it can be to develop a coding scheme based on the functions that the statements have.

Let us examine further how scientists create coding schemes and how we can refine the schemes for our examples. The first question we should ask about a coding scheme involves how we can describe the topic. We need to ask whether we should base the classification system on a continuous or a categorical description of the behavior that we are analyzing. As its name implies, categorical description classifies behaviors according to a set of categories. There are certain behaviors that we cannot order on a continuum, such as one moving between "more" and "less." Therefore, we must analyze these behaviors according to a categorical scheme instead.

As with most verbal interaction schemes, the system that Bales created is categorical. Different functions for statements include "shows solidarity" or "shows tension release." We cannot order these functions in relation to one another on any dimension. Thus, we cannot place them on a continuum. Instead, we must use a categorical coding scheme for these statement functions.

There are behaviors, however, that we are able to order on a continuum. We can analyze such behaviors according to either a categorical or a continuous scheme. For instance, a coding pattern that merely separates a member's silences from his or her "sounds" is categorical. On the other hand, a classification system that describes the loudness of the member's voice at different times, according to a numeric scale, would be continuous. Continuous descriptions are more informative than categorical ones. However, they are also more difficult to use. Due to this difficulty, from now on we will restrict our discussion to categorical descriptions.

Coding Scheme Requirements

There are certain requirements for a good categorical coding scheme. First, it needs to be as simple as possible. It should be so, however, without sacrificing the significant distinctions the researcher intends it to address. Coding schemes are simple if they have relatively few categories. This is because few categories are relatively easy for researchers to learn and to use. Classification systems are also simple if scientists define their categories clearly. Clear definitions allow users to know exactly which behaviors do and do not fit each category. For our example, "sound" versus "silence" is as simple a coding scheme as a researcher could have. Bales's system has a relatively large number of categories. Nevertheless, his organizational scheme eases its difficulty of use.

Content analytic researchers also say that coding schemes have two additional requirements. The first of these is that categories are supposed to be exhaustive. This means that every behavior fits into one category or another. The second requirement is that the categories are mutually exclusive. In order to be so, every behavior should fit into only one category, never into two or more. However, there are ways of getting around both of these additional requirements. For instance, some coding schemes allow sieve coding. This means that some statements are not coded within the system. We described some of these in chapter 8 when we showed schemes that ignore maintenance-oriented statements.

There are other ways in which coding schemes can avoid these two additional requirements. For example, some systems allow multifunctional coding. With these schemes, a researcher can put a statement into more than one category. The best way to do this, however, is to maintain exclusivity in another way. For instance, a scientist could put statements that serve function "A" in one category, statements that serve function "B" in a second category, and those that serve both "A" and "B" in a third category.

There is a problem with such an option, however. The problem is that as the number of categories in the scheme increases arithmetically, the number of possible combinations increases exponentially. The following table shows how this would happen:

Number of categories	1	2	3	4	5	6	7	8
Number of combinations	1	3	7	15	31	63	127	255

(These combinations include each category alone.)

For example, with Bales's 12-category scheme, the number of possible combinations would be 4,095! However, there are times when such an option might be feasible. For example, if only a few of the possible combinations actually occur in conversation, the procedure would be manageable. Hewes (1979) discussed these issues.

Validity and Reliability

Once a scheme has met the basic requirements discussed above, researchers must evaluate the system further according to additional criteria. They must evaluate the coding system for its validity and its reliability.

Validity. A scheme is valid if it measures what the researcher intends it to measure. There are various types of validity. Researchers say that a coding system has face or content validity if it makes intuitive sense to its users and to others who evaluate it. In order to make sense, a scheme should define its categories clearly, and these categories should measure what the researcher intends them to measure. A system has construct validity if the categorization of behaviors within it is consistent with a theory that the system intends to test. Kerlinger (1964) discussed face and construct validity.

A coding system may also have representational validity if it makes the same types of distinctions among categories that the average person would make (Poole & Folger, 1981a). In other words, the average person and the coding scheme would agree on how to classify particular behaviors.

Any coding scheme requires face validity. However, the other kinds of validity apply to specific types of coding schemes. For instance, if a classification system measures behaviors that people naturally interpret, it needs to have representational validity. Another instance is that any scheme that is consistent with a specific theoretical perspective must have construct validity.

Reliability. In common usage, reliability is a synonym for such words as "dependability," "stability," and "consistency." This is also true of the way in which researchers use the word "reliability" when they apply it to coding schemes. A scheme is reliable if one coder can categorize a behavior the same as a second coder. For example, Bob and Marcie watch the same group and code the interaction in the same way. Further, a scheme is reliable if one coder can categorize the same behavior the same way at two different times. For example, Bob sees a videotape of a group four times, and each time he codes the interaction in the same way.

A classification system needs to have clarity and ease of use in order to be reliable. Those who use the system must also have a sufficient amount of training and practice in order for the system to be reliable.

Guetzkow (1950) distinguished between two types of reliability that a coding scheme must have. The first type he distinguished was categorizing reliability. This term means that two coders, or the same coder at different times, can place the same statement in the same category. Guetzkow's other type of reliability was unitizing reliability. This means that two coders, or the same coder at different times, agree on the "size" of the behavior. By the word "size," Guetzkow meant where the behavior begins and where it ends, and the like.

For instance, when analyzing verbal interaction, units of measurement can be as small as phrases and sentences or as large as an entire conversational "turn." A "turn" lasts from when a person starts to speak until the time he or she finishes. If two coders do not agree on the size of the unit, their separate codings of the same conversation will yield vastly different results. For instance, the smaller the unit, the more acts the coder will code. Therefore, different unit sizes affect how many acts each coder will analyze.

Even if coders agree on unit size, there can be problems if they do not agree on the "time" that each act should take. For example, it is possible that one coder will see acts beginning at Times 1, 3, 5, and 7. In contrast, another coder will see acts starting at Times 2, 4, 6, and 8. The two coders will come up with the same number of acts, but they are not coding the same acts, and difficulties arise.

Thus, in order to be successful, behavioral coding requires that coders be reliable both in their categorization and in their unitizing. This reliability, by necessity, presupposes a good coding scheme, good training of the coders, and sufficient practice together by the coders. These factors ensure that the researchers are coding similarly.

Time, Data, and Sampling Intervals

Given a valid and reliable coding scheme, there are further criteria regarding the scheme. The next set of questions concerns how scientists will use the coding system.

Time. The first of these queries regards what type of time the researchers will use. Will they code the behaviors in event time or in clock time? If they use event time, they will note only the occurrence of a type of behavior within its category. For instance, if Jane speaks and Sue is silent or if Jane "asks for information," an event has occurred.

In contrast, clock time coding involves two notations. Using clock time, an researcher will note the occurrence of a type of behavior and also the length of time, on a clock, during which the behavior occurs. Clock time is particularly useful when a person is studying nonverbal interaction. An event time coding of sound-silence patterns between two people may tell little more than that the two people take turns talking. However, knowing the duration that each person talks can tell us more. For example, it can tell us which of the two people dominated the discussion by talking longer.

As we pointed out earlier, there is usually a trade-off between the amount of information that a certain technique can reveal and the difficulty in using that technique. Such a trade-off exists in this case, too. Clock time coding does lead to more information than event time. In addition, a researcher can deduce all event time information from clock time data, whereas the reverse is not true. However, clock time coding is more difficult than event time to implement.

Data. The next question regarding how the researcher will use the scheme concerns the type of data the scheme requires. Should the researcher collect static or dynamic data?

Static interaction data ignores the "structure" of the interaction. It focuses only on the number of acts within each category. For example, let us suppose we are studying the sound-silence patterns in a discussion between Jane and Sue. We will have four categories in our sound-silence coding scheme: Jane talking/Sue talking, Jane silent/Sue silent, Jane talking/Sue silent, and Jane silent/Sue talking. If we collect static data, we note when an occurrence of each category takes place. We could end up with a table such as the following:

J talk/S talk	J talk/S silent	J silent/S talk	J silent/S silent
5	5	5	5

From this table of data, we learn that there are an even number of events in each category. There are five events in which each member is talking and the other is silent, and five when both are talking. In addition, there are five events when both are silent.

However, a dynamic analysis considers the interaction process of the conversation. We discussed this topic in Chapter 8. To perform a dynamic analysis, we must note the number of times that the acts follow one another during the conversation. We might get the following results from an event time analysis:

			From
		J talk/S talk	J talk/S silent	J silent/S talk	J silent/S silent
	J talk/S talk	-	5	-	-
To	S talk/S silent	-	-	-	5
	J silent/S talk	5	-	-	-
	J silent/ S silent	-	-	5	-

From this data we can see a definite and consistent pattern to the interaction. When Jane is talking, Sue eventually interrupts. Whenever this happens, Jane falls silent until Sue finishes speaking. There is then a joint silence. Only then will Jane speak again. It appears that Sue is dominating the conversation.

We have here a rudimentary "process" for the conversation. Jane and Sue appear to follow unspoken rules. Dynamic data thus allows for certain types of sequential analysis. Further, a researcher can combine this analysis with clock time data. Such a combination may, for example, show that Sue has longer speaking turns than Jane does. This is more evidence that Sue dominates the discussion.

Sampling intervals. There is one last factor that a scientist must consider when creating a coding scheme. This problem is more apparent in continuous clock time analysis. It is the issue of sampling intervals. Simply put, the scientist needs to decide how often to "look in" on the conversation in order to note what is taking place. Let us imagine that we have decided to study back-channel utterances. This would occur, for example, if Jane were talking, Sue were to interject a short supporting statement such as "uh-huh," and then Jane were to continue talking. Saying "uh-huh" takes about half a second. As you can calculate, if we sample the conversation at one-second intervals, we run into problems. We will miss many occurrences of "uh-huh." They will slip through while we are not analyzing the conversation. If this happens, our results will bias us toward believing that the conversation was smoother than it actually was.

In contrast, we could sample more often than at one-second intervals. For instance, we could record data every eighth of a second. We will then get all of the "uh-huhs." However, it is possible that we will now need to code verbal behaviors of extremely short duration, such as coughs. Our results may now be biased in the opposite direction. This will lead us to believe that the conversation was less smooth than it actually was. One of the big problems with continuous clock time coding is deciding on the sampling interval that is best. Researchers must compromise between their desire for adequate detail and their wish to ignore inconsequential behavior.

After you have answered all the relevant concerns above, you have created an important tool, a coding scheme, to help you know "how" to study your topic. Your decisions and preparations have made you ready to code a group interaction. This is a difficult task in and of itself, requiring intense concentration to ensure that the researcher does not miss something important. It is hardest for an researcher to perform when he or she must observe a live meeting. If the coder misses a behavior, he or she cannot retrieve that data for analysis.

Thus, it is advantageous to make a permanent record of a live meeting on tape. The coder can then review the discussion at a later time. Videotape is best. In one medium it offers both a visual and a verbal record of the group's activity. If audiotape is all that is available, the researcher loses a record of the visual activity and consequently must code any relevant nonverbal actions on the spot during the live meeting. It is also very difficult to note who said what, when listening to an audiotape, if each group member does not have a distinct voice. This problem does not occur when a person codes from videotape because he or she can see who is talking.

Thus, if an researcher can only use an audiotape, she or he needs to take careful notes during the live meeting. It is necessary that the researcher keep a written record of who is saying what during the conversation. A person can do this by listing the group members who talk in the order of their contributions. In addition, the person can write key words or phrases regarding what members say next to their names. This allows the researcher to go back during later analysis and match the written record with what the group members say on the audiotape.

Whether a scientist uses a videotape or an audiotape, he or she should be sure to check the reliability of the coding analysis. Preferably, a person does this by comparing his or her coding of a conversation with another researcher's coding. If an researcher is working alone, he or she can make two codings of the same conversation at different times and then compare these two codings.

Once a researcher has collected her or his data, there are further steps involved in "how to study" group process. Interpretation is the step that follows data collection. If a researcher has been careful in linking behavioral indicators to the features he or she is interested in, interpretation should be comparatively easy.

However, researchers are sometimes prone to make bolder interpretations than they reasonably should. This is particularly a problem with nonverbal behavior. Part of the reason that many people misinterpret nonverbal behavior is a rash of books that appeared during the 1970s. These books oversimplified the significance of nonverbal actions. They did so in order to entice people to buy a book that publishers guaranteed would teach readers what their own and other people's behaviors "really mean."

The problem is that nonverbal behavior does not "mean" anything in the usual sense of the word. For instance, we might say that a noun "means" an object to which it refers. Nonverbal actions do not function in this way. A person cannot write a "dictionary" explaining nonverbal actions. For example, one of these books claimed that, if a person sits with arms folded tightly around himself or herself, he or she is not interested in the person of the opposite sex nearby. This book ignored the idea that this behavior could mean other things as well. For instance, tightly folded arms could indicate that the person is cold or perhaps has a stomachache. There are many reasons for such a nonverbal action.

There are ways to aid a researcher so that he or she does not misinterpret nonverbal behavior. The trick to interpreting such behavior is to put the action in the context in which a person performs it. This is because its meaning is only within that context. Knapp (1978, p. 381) provides a list of the types of things that an researcher should consider when he or she interprets nonverbal behavior:

1. The researcher should consider the "group members" themselves. This includes looking at their age, sex, position, status, relationship to one another, previous history as a group, and so on.

2. The researcher should consider the "setting" or kind of environment in which the interaction takes place. For example, is it a home, public building, park? Also, what is the relationship of the members to the environment? For instance, does a group member live in the house, or is he or she visiting? The setting further includes the behavior that people expect in the environment. For example, libraries and bars imply different behaviors.

3. The researcher should consider the "purposes" of the interaction. These include the goals of the group, goals of each member, the compatibility of these goals, and so on.

4. The researcher should consider the "implications" of the behavior. What initiated the action? What are its effects on the other group members? This also includes the extent to which the behavior is normal or rare for a particular person to perform in a particular situation.

Now we have covered all the basic topics that a researcher needs to consider as he or she prepares to examine group process. Before we move on to the rest of the appendix, we would like to say a final word on research ethics. People have a right to know when someone is studying their behavior. Researchers should inform everyone that they observe. This is true even if the people under observation would never have learned about the research study unless the researcher told them. A person should never do a behavioral study without getting prior permission from the people participating in the study.

In particular, researchers should inform research participants when the researchers are making videotape or audiotape records. Once something is on record, there is a chance, no matter how much care a researcher takes, that the evidence of a behavior can hurt a person. For example, a researcher could videotape a group of college students socializing. John, a member of the football team, is part of the group. During the videotaping session, John takes a drink from a friend's beer. If his college has stiff penalties for alcohol use during the football season, evidence of John's behavior could hurt him, no matter how careful the researcher is.

There is always a conflict between ethics and situations in which researchers feel that they want test participants to act "naturally." There is informational value in research with participants who are naive to the study. However, luckily for researchers, it appears that once people get used to a camera or microphone, they begin to act naturally even when someone records their behavior. Wiemann (1981) discussed this phenomenon. Further, there are situations scientists call "naturalistic behavior" studies. An example of this type of study is the ever-popular study of pick-up behavior in bars. When a researcher conducts such an investigation, it is best to be up-front. The researcher should not hide his or her notebook and stopwatch. In addition, the researcher should answer all questions about his or her business honestly.

Overall, most people do not mind being observed for a good cause. They will object, however, if they feel that the researcher is taking advantage of them. That is why it is a good idea to observe people in an obvious manner, being honest and up-front about the study. If a researcher does this, it is not likely that the people under observation will feel abused. However, if the observer is "caught" trying to study people on the sly, those under observation may very well feel animosity toward the researcher.

Throughout this book, we have discussed a series of topics, such as cohesiveness, process, and leadership, that can be studied through content analysis. In the remainder of this appendix, we will describe a number of behavioral indicators for each. In some cases, the researcher can use published coding schemes for nonverbal and verbal interaction. We can assume that scientists have found these schemes to have at least a reasonable reliability rating. In addition, we can believe that their users have considered them to have face validity. However, we should not consider that these coding systems have construct or representational validity, unless we specifically state that they do.

There are a number of different verbal and nonverbal behaviors that can be used as indicators of group cohesiveness. One of the best verbal indicators is the "We/I ratio." To compute the We/I ratio, the researcher computes the number of times that group members use pronouns such as "we," "us," and "our" to refer to the group and compare this number with the number of times that group members use pronouns such as "I," "me," and "mine" to refer to themselves. The larger the proportion of "group reference" pronouns to "personal reference" pronouns, the more evidence the researcher has that the group is cohesive. The We/I ratio was used as an indicator of cohesiveness in the critical studies on leadership style performed Lewin and his associates, as described in Chapter 10.

Another good verbal indicator of cohesiveness are the maintenance functions in Bales's coding scheme as discussed in Chapter 8. In this case, one would compare the number of utterances coded as positive maintenance ("shows solidarity," "shows tension release," and "shows agreement") with the number of utterances coded as negative maintenance ("shows antagonism," shows tension," and "shows disagreement"). Bales used this comparison in his own research, as discussed in Chapter 8.

Turning to nonverbal indicators, the researcher should consider the "immediacy" cues recognized by Mehrabian (1973; see Chapter 9). Immediacy cues are indicators of liking. Many of these have been validated researchers as cohesiveness cues. For example, Piper, Marracke, Lacroix, Richardson, and Jones (1983) found that a group member's commitment to the group correlated with the average distance between that person and the other group members. In other words, closer distance means more commitment. Tickle-Dugnan and Rosenthal (1987)'s review of relevant studies revealed liking to be related with forward trunk lean, direct body orientation, open arm position, mirroring of one another's posture, and to a lesser extent uncrossed legs. Other immediacy cues that may be related to group cohesiveness include touching, eye gaze, and smiling.

When studying nonverbal indicators of cohesiveness, it is just as important to code indicators of noncohesiveness. Do not forget to note instances of, for example, leaning away and gaze aversion.

Conflict is usually best studied through verbal indicators. When studying conflict, it is important to look for behaviors indicating both the presence and absence of conflict. For example, the researcher might analyze the number of statements implying both agreement and disagreement among group members. One should also look for instances when group member are voicing support for competing viewpoints or proposals and contrast those with examples of group members voicing support for the same position. Ellis & Fisher (1975) studied group conflict in these ways, using Fisher's coding scheme to be described below.

We are not aware of any coding schemes specifically designed to study group conflict. There are schemes for studying interpersonal conflict that might be adapted to the group discussion. One such scheme was developed by Sillars, Coletti, Parry, & Rogers (1982). It includes 27 categories grouped into three major types; utterances that imply the avoidance of conflict, utterances that imply competition, and utterances that imply cooperation.

Theorists have proposed several coding schemes to help researchers analyze the process of negotiation. We will limit our discussion to the most sophisticated of these schemes.

Morley and Stephenson scheme. Morley and Stephenson (1977) offered one such coding system. Their idea was based on Longabaugh's (1963) work. Before we discuss Morley and Stephenson's system, we will review Longabaugh's work.

Longabaugh worked within the tradition of social exchange theory (see Chapter 5). According to Longabaugh, there are certain requirements for a coding scheme when the scheme is based on social exchange theory. Such a coding system needs to distinguish between two "dimensions" of an utterance. The first dimension is the mode of the statement, or the method of exchange. Was the person seeking, offering, taking away, withholding, accepting, ignoring, or rejecting?

The second dimension involves the resources, or commodities, that the statement exchanges. Unfortunately, Longabaugh did not propose a set of resources that scientists could universally apply to statements. In his own research, he looked for the resources of information, freedom, and support. However, Longabaugh also admitted that an indefinite number of resource types may exist. This claim negates the face validity of his scheme. Researchers cannot use a coding scheme that includes a dimension with an infinite number of categories.

Morley and Stephenson's scheme was an offshoot of Longabaugh's work. Their scheme, called "Conference Process Analysis," does not have this problem with face validity. It distinguishes four modes: accept, reject, seek, and offer. The scheme also delineates nine types of resources: procedure, settlement point, limit, positive and negative consequences of proposal, other statements about outcome, positive and negative acknowledgment, and information. In addition, Morley and Stephenson proposed a third dimension, involving types of reference. These include none, self, person, other, party, opponent, both persons, and both parties. The scheme has a degree of construct validity because it is based on the social exchange point of view.

Hopmann and Walcott scheme. Hopmann and Walcott presented another coding system for negotiation. Schelling (1960) had conducted work involving types of conflict behavior. His work inspired Hopmann and Walcott's coding scheme.

The Hopmann and Walcott scheme distinguishes six major functions of statements made during negotiation. The first of these functions is "substantive" and involves messages that facilitate the negotiation process. The second is "strategic" and includes statements that are designed to influence the opponent. The third function is "persuasive" and refers to messages that function as part of arguments concerning proposals. The fourth is "task" and refers to messages that promote a "businesslike discussion" of the issues. The fifth function is "affective" and involves statements that express feelings about the issues and the opponent. The sixth is "procedural" and refers to statements that focus on organizing the discussion.

Hopmann and Walcott subdivided each of these function categories into specific kinds of statements. Putnam and Jones (1982) used this scheme in their study (see Chapter 4) and describe Hopmann and Walcott's coding system well in their report.

Donohue scheme. Donohue (1981a) presented another coding scheme related to negotiation. He based his conflict research (1981b) on this scheme (see Chapter 4). Donohue's scheme is consistent with the interactional perspective, which categorizes each utterance in relation to the statements that immediately precede and follow it. Hence, the scheme codes each utterance twice. The first coding is in terms of the statement's function as a response to the preceding utterance; the second coding involves the statement's function as a cue for the subsequent utterance.

For both types of coding, the scheme examines the statement for its "grammatical form," for the "tactic" that it represents, and for its "weight" or intensity. The grammatical form could be a talk-over, question, assertion, or incomplete form. The tactic could be attacking, defending, or regressing, which is the same as giving in to something. The weight of an utterance ranges from "five" for the most intense attack to "one" for the most intense regression.

We have described nonverbal behavioral indicators of power, and verbal indicators of powerlessness, in some depth in Chapter 5. Perhaps the most important of these is talk time. Powerful people talk more than the powerless, and the number of utterances or actual time spent talking by each member is easy to measure.

Similar to the concepts of "power" and "powerlessness" are the concepts of "dominance" and "submission." Researchers from the interactional perspective have developed coding schemes for studying dominance and submission, based on the sequence of utterances. We will describe the two best-known of these methods.

Ericson and Rogers scheme. Ericson and Rogers (1973) proposed a coding system that is consistent with the interactional perspective. Millar and Rogers (1976) described this scheme well. The classification system from Ericson and Rogers analyzes how conversation indicates the patterns of dominance and submission in an interpersonal relationship. Their scheme codes each statement for "form" and "function." Form includes assertion, question, talk-over, incomplete, or other. Function involves support, nonsupport, extension, answer, instruction, order, disconfirmation, topic change, initiation/termination, or other. There are a total of 5 X 10, or 50, possible combinations of forms and functions in this scheme.

Each of these 50 categories indicates how a discussion statement relates to dominance or submission. When a speaker attempts to define the interaction, or dominate, the scheme codes this as a "one-up move." The researcher would designate this kind of statement with an arrow pointing upward. If a person instead accepts another's definition of interaction, or submits, this is a "one-down move." The researcher indicates this with an arrow pointing downward. When a discussion move is irrelevant to dominance or submission, the scheme codes it as a "one-across move." The coder designates such a move with an arrow pointing to the right.

For example, the system codes most talk-overs as bids for dominance. The exceptions to this are when the talk-overs function as support for the listener or when they are in the miscellaneous "other" category. The scheme generally considers other forms also as bids for dominance, unless they function as support or as extensions or unless they are in the "other" category.

Hence, the researcher would code each statement as a "one-up," a "one-down," or a "one-across" move. The coder could then, for example, count a person's moves to determine if he or she generally attempted to dominate or submit during the conversation. However, such a count would not show whether the person actually succeeded in dominating or submitting during the discussion. To evaluate the speaker's success, the coder needs to analyze the "interact structure" of the conversation. In particular, the researcher should note those interacts that have only one-up and/or one-down moves.

Interacts that consist of a one-up move and a one-down move are complementary. They indicate that, at least in that part of the conversation, there is agreement on who dominates and who submits. Interacts that have two one-up moves are examples of competitive symmetry, which are battles for dominance. Interacts consisting of two one-down moves are examples of submissive symmetry. They are battles for submission.

Through a dynamic analysis, the coder can follow dominance or submission battles over the length of the conversation. The researcher can thereby see whether the participants resolved these battles and, if so, by what manner. The coder can also determine who controlled the conversation. In other words, he or she can find out whose bids the participants accepted most often. We should not confuse "control" with "dominance." A person can control a discussion even when he or she apparently is set on taking a submissive role.

Ellis scheme. Ellis (1979) proposed another scheme for measuring dominance and submission. Fisher and Drecksel (1983) described the Ellis coding system best. Ellis distinguishes five "types" of moves:

1. Domineering. This is a strong one-up move, or attempt to restrict the listener's conversational options. The symbol for the strong one-up move is an arrow facing upward followed by a "+" sign.

2. Structuring. The structuring type of move is a weak one-up, or attempt to lead the conversation without restricting the listener's options. The weak one-up move has the symbol of an arrow facing upward followed by a"-" sign.

3. Equivalence. This is a one-across move, as before.

4. Deferring. The deferring move is a weak one-down, or an attempt to "follow" the listener's direction conversationally. The symbol for a weak one-down is an arrow facing downward followed by a "-" sign.

5. Submitting. This is a strong one-down move, or an attempt to relinquish all conversational options to the listener. The strong one-down move is symbolized by an arrow pointing downward followed by a "+" sign.

Folger and Sillars (1980) studied the representational validity of both the Ellis scheme and the Ericson and Rogers scheme. This study by Folger and Sillars had mixed results. The Ericson and Rogers coding scheme was often consistent with the ways in which the participants evaluated the degree of dominance or submission that statements implied. However, there were several exceptions to this consistency. The most significant departure from the way in which participants judged statements involved the way that Ericson and Rogers coded questions.

Folger and Sillars also found mixed results regarding the Ellis classification system. They found support for his distinction between strong and weak one-up statements. However, they did not discover support for his analogous distinction between strong and weak one-down utterances.

O'Donnell-Trujillo (1981) coded 18 discussions between husbands and wives for dominance and submission using both the Ericson and Rogers scheme and the Ellis scheme. He wanted to find out if the two schemes classified the same statements in the same way. For example, would a statement that would be coded as a one-up using the Ericson and Rogers scheme also be coded as a one-up using the Ellis scheme? His results were disappointing. The correspondence was not great. This finding presents a challenge to the face validity of both coding schemes.

Canary, Ratledge, and Seibold scheme. Canary, Ratledge, and Seibold (1982) developed a scheme for the study of group argumentation. Canary, Brossman, and Seibold (1987) describe it well. The scheme classifies statements into four categories: arguables, prompters, delimiters, and miscellaneous, or non-arguables. The first three categories are further classified into a few specific subtypes.

When using the scheme from Canary et al., it is best to perform coding by using a transcript of the discussion. A researcher should give each speaking "turn" a sequential number, starting with "one." The researcher then should include each number with the function of the statement. Sometimes, a speaking turn will include statements that belong in other categories. In this case, the coder should number each of these subparts sequentially.

For example, "turn 14" might include the subparts of "14.1 assertion," "14.2 elaboration," "14.3 justification," and so on. When a researcher numbers statements, he or she can study how arguments disappear and reappear throughout a discussion. Certain qualities should aid the researcher as he or she studies these argument "spirals." For example, the coder can mark statements after which a line of thought "disappears" with the number of the utterance when it "reappears." Next to a statement, he or she might write, for instance, "see turn 68.2." Researchers can use parentheses to indicate the beginning and end of associated statements.

Hoffman scheme. Hoffman (1979) created a very simple scheme to examine his valence model of decision development (see Chapter 7). He coded utterances toward particular proposals as "favorable," "unfavorable," or "neutral." His scheme is derived from this model and, therefore, we can say that his coding system has construct validity. McPhee, Poole, and Seibold (1981) critiqued Hoffman's work (see discussion in Chapter 7). In their critique, McPhee et al. note that Hoffman's scheme is insufficiently justified. In reaction, they proposed their own system for coding valence, which is also derived from a valence model. Thus, it has construct validity. They also claim evidence of representational validity for their system.

Vinokur, Trope, and Burnstein scheme. The researcher could use Hoffman's scheme to study group polarization, but Vinokur, Trope, and Burnstein (1975) proposed a more complex scheme for that purpose. The Vinokur et al. scheme borrows concepts from the subjective expected utility model discussed in Chapter 12. Each utterance is coded as relevant to either:

1. Utilities of success and failure for either the risky or cautious choice.

2. Utilities inherent in either the risky or cautious choice other than success and failure.

3. Probability of success of the risky or cautious choice.

4. Explicit recommendation for taking either the risky or cautious choice.

5. A miscellaneous category for other utterances.

These are the coding schemes used in some of the important studies of group process discussed in Chapter 8.

Bales's Interaction Process Analysis. Unquestionably, the most important verbal coding scheme of any type and the one that researchers know best is Bales's Interaction Process Analysis (IPA), developed in 1950 and discussed in Chapter 8. As you recall, Bales deduced the categories of IPA from his functional theory. Thus, IPA is one of the few coding schemes with construct validity because its structure follows from theory.

Through the 1950s and 1960s, researchers used IPA extensively for the study of verbal interaction. Many of these studies, however, were descriptive and nontheoretical. These descriptive studies used IPA independently of Bales's theory. For example, IPA was used in many studies of negotiation, although there is no reason to believe that its categories are relevant to that circumstance. Under these circumstances, the coding scheme had no construct validity. Thus, if an researcher were to use IPA without simultaneously adopting Bales's theory, he or she could not claim construct validity for that particular use of the scheme. Poole and Folger (1981b) further found that Bales's coding scheme has good representational validity, allowing Poole to use IPA in his tests of the phase hypothesis (1981, 1983a), described in Chapter 8.

Scheidel and Crowell scheme. Scheidel and Crowell (1964) proposed a coding scheme that they intended would help researchers follow the development of group proposals. Unlike IPA, Scheidel and Crowell's scheme is like most classification systems for small group communication in that it ignores maintenance comments. Therefore, it requires "sieve coding." We described Scheidel and Crowell's scheme in Chapter 8, along with their research findings.

Fisher scheme. B. A. Fisher (1974) described a similar coding scheme, for the same purpose regarding group proposals. His system has the following categories:

1 - Interpretation	2 - Substantiation
f - favorable	f - favorable
u - unfavorable	u - unfavorable
ab - ambiguous	ab - ambiguous
an -neutral	an - neutral
3 - Clarification	4 - Modification
5 - Agreement	6 - Disagreement

Poole and Folger (1981b) found that Fisher's scheme has reasonable representational validity. Poole also used Fisher's categories in some of his studies (1981, 1983a). In Chapter 8, we described Baird's (1974) coding scheme, which is, despite its simplicity, comparable to Fisher's.

Hirokawa scheme. We have discussed many of the studies in Hirokawa's research program in Chapters 8 and 12. In some of these studies, he used versions of a coding scheme best described in his 1982 essay. He called this coding scheme the Functional-Oriented Interaction Analysis System. The coder using Hirokawa's scheme first classifies speaking turns in the following categories:

1. Establish operating procedures

2. Analysis of problem

3. Generation of solutions

4. Evaluation of solutions

The coder then further classifies speaking turns according to a 12-category scheme similar to Fisher's.

Poole schemes. The Poole and Roth (1989a, 1989b) studies, described in Chapter 8, analyzed group discussion according to two schemes: task and maintenance. The task-oriented Decision Functions Coding Scheme is based on both Bales's and Fisher's schemes. Statements are coded according to the following categories:

1a - Problem analysis	3c - Solution elaboration
1b - Problem critique	3d - Solution evaluation
2a - Orientation	3e - Solution confirmation
2b - Process analysis	4 - Tangent
3a - Solution analysis	5 - Simple agreement
3b - Solution design	6 - Simple disagreement

Contiguous statements are then organized into interacts, and strings of interacts are then classified according to the following "phase indicators":

PA - Problem Analysis	OO - Orientation
PC - Problem Critique	NN - Tangent
SA - Solution Analysis	CF - Confirmation
SD - Solution Development	DIS - Disorganized Period
SC - Solution Critique

For example, a series of acts such as 2a, 2a, 2a would be coded as an OO phase indicator; a series of acts such as 3a, 3b, 3b, 3c would be coded as an SD phase indicator. The sequence of phase indicators is then further analyzed to find task-oriented discussion phases.

The maintenance-oriented Group Working Relationships Coding System includes the following categories:

1 - Focused work	3c - Tabling
2 - Critical work	3d - Open discussion
3a - Opposition	4 - Integration
3b - Accommodation

The coder then divides the discussion into 30-second stretches to see which categories predominate during that period. Next the coder assigns the code for that predominant category to that entire 30-second segment. Thus, if a given 30-second segment consists mostly of statements coded 3d, the coder considers that an Open Discussion segment. The sequence of segments is then further analyzed to find maintenance-oriented discussion phases.

In Chapter 9 we discussed research relevant to situations when a structure of who talks to whom, such as a circle or wheel, is imposed on a group. Most group discussion occurs without an imposed structure. However, structures of who talks to whom tend to emerge in these situations. For example, the group's most talkative member receives a large proportion of utterances from the other members. Bales (1953) is an example of a researcher who has performed this type of study.

To examine emergent group structure, the researcher counts how many utterances each member of the group makes and who the utterances are directed to. In this way, the researcher can see if, for example, one member both makes and receives most of the utterances (an emergent wheel-like structure), or if all members makes and receive about the same amount (an emergent comcon). One thing to keep in mind is that some utterances may be directed toward more than one member, or toward the group as a whole. This fact complicates the analysis.

Given that the leader tends to do most of the talking in the group, one way to measure leadership is through talk time, as Bales (1953) did. However, this type of measurement is probably better for studying power than for leadership. A better way to study leadership is through the functional approach. The researcher codes every member's utterances that explicitly provide either procedural, substantive, or maintenance leadership. This allows the researcher to make very specific claims about how much and what type of leadership each group member is providing for their group.

Pavitt, Whitchurch, McClurg and Petersen scheme. In order to perform our study described in Chapter 11 (Pavitt et al., 1995), we developed a "leadership functions coding scheme." The scheme included leadership functions that research participants had judged as critical to leadership, and was also influenced by the list of group functions displayed in Chapter 8. The scheme includes six procedural functions (procedure developing, orienting, agreement-testing, expediting, energizing, and assigning), four substantive functions (summarizing, evaluating, seeking, and mediating), and three maintenance functions (conflict managing, encouraging, and harmonizing). Other utterances are placed either in nonleadership procedural, nonleadership substantive, or nonleadership maintenance categories.

Hewes, Planalp, and Streibel scheme. Hewes, Planalp, and Streibel (1980) proposed a method to evaluate how a group uses its discussion procedure. To begin their method, they taught groups a version of Reflective Thinking. Scheidel and Crowell (1979) created this particular version and called it "Problem Management Sequence." Hewes et al. then evaluated how the groups used this version of Reflective Thinking. To do so, they coded the group's discussions according to a scheme. This scheme consists of the major steps in the Problem Management Sequence. The coding system is as follows:

1. Clarification of problem

2. Specification of symptoms

3. Appraisal of scope

4. Identification of causes

5. Establishment of criteria

6. Presentation of solutions

7. Evaluation of solutions

To have an exhaustive scheme, Hewes et al. added three additional categories. These were (1) simultaneous presentation/evaluation of solution, or misuse of the procedure, (2) maintenance statements, and (3) procedural statements.

Hewes et al. examined the statements that groups made and matched them to the seven categories listed above. To the extent that these statements occurred in the same numeric order as the seven categories, the groups were following the procedural steps of the Problem Management Sequence in the correct order. The researchers found that training helped the groups use the procedure more correctly. This finding was due to the fact that the group discussions more closely approximated the numeric order of the categories after training than before it.

This general strategy for devising a coding scheme for studying group procedure can be applied in other ways. A researcher can take the steps in any formal procedure and use them as the categories in a coding scheme. Then the researcher can examine how closely a group's procedure approximates the scheme. If a practitioner believes that a particular formal procedure represents the "ideal" for a group in a particular situation, the practitioner can then analyze the extent to which the group's discussion approximates that "ideal."

We hope that this survey of coding schemes and other behavioral indicators is helpful. As you can note, researchers have many different options when they analyze group process.