Changing the Subject: Gender, Status, and the Dynamics of Topic Change

by Dina G. Okamoto, Lynn Smith-Lovin
Changing the Subject: Gender, Status, and the Dynamics of Topic Change
Dina G. Okamoto, Lynn Smith-Lovin
American Sociological Review
Start Page: 
End Page: 
Select license: 
Select License



University of California, Davis University of Arizona

Social scientists have devoted a great deal of attention to how much people talk, but have paid little attention to what they talk about. Research in the tradition of con- versation analysis suggests that transitions between topics of conversation are ac- complished in a systematic, structured way, and that social status can affect whose topics are developed and whose are lost. The authors use insights from conversation analysis to develop a systematic coding system for identifiing topic shifts in task- oriented discussions. Hypotheses from the literature on group processes predict who will suggest topic changes in a task-oriented group and whose topics will be lost. Event history methods model the dynamics of topic change in two data sets: a study of six-person laboratoiy task groups and a replication study of dyads. Topic changes in these task-oriented discussions are more sensitive to status structures that develop within the conversation than to a relatively weak status characteristic like gender. Some of the sequential mechanisms that conversation analysts have studied in the context of less structured, more wide ranging talk may be generalizable to this more

constrained conversational environment.

Please address correspondence to Dina G. Okamoto, Department of Sociology, University of California-Davis, 1282 Social Sciences and Humanities Building, Davis, CA 95616 ( The authors would like to thank Paula England, Miller McPherson, Linda Molm, Dawn Robinson, and four anony- mous reviewers for comments on an earlier ver- sion of this paper. This research was supported by an National Science Foundation Grant SBR- 941 1157 to Lynn Smith-Lovin.

Some patterns found initially have become

Turn-taking patterns have important effects on people's abilities to express ideas, have their concerns addressed, and receive credit for their contributions. Settings like marital disputes, doctor-patient consultations, and work groups are domains where the ability to have one's say and get attention for one's ideas can be vitally important for individual and collective outcomes.

The model of turn-taking (Sacks et al. 1974) created a useful synergy between scholars who analyze the structures of con- versation with qualitative methods and the more quantitative tradition that focuses on status-organizing processes in small task groups. Traditionally, group processes re- searchers studied participation, evaluative

less clear as a larger body of research has devel- oped (see reviews by Aries 1996; James and Clarke 1988; Thorne, Kramarae, and Henley 1983). It now appears that status, power, institu- tional contexts and local conversational dynam- ics are more important than social categories like gender in structuring conversational behavior.

statements, influence, and assertive gestures in order to find out how status hierarchies emerged in groups (for a review, see Ridge- way and Walker 1995). In the last 10 years, research by conversation analysts led group processes researchers to add a focus on in- terruptions, overlaps, back-channel cues, and other turn-taking structures (Johnson 1994; Johnson, Clay-Warner, and Funk 1996; Robinson and Smith-Lovin 1990; Smith-Lovin and Brody 1989; Smith-Lovin and Robinson 1992). The argument was that deviations from turn-taking norms (e.g., in- terruptions) would be determined by the status structures within a group.

The consistent focus on the structure of talk in task groups has led to an interesting oversight: little attention has been paid to what is being talked about-the topic of the discussion. Researchers typically break the group conversation down into small units (e.g., action opportunities or utterances) or aggregate it into larger, nonstructured units (e.g., total participation). Researchers pre- sume that these structures tap the impact that different speakers have on the develop- ment of the group's potential task solutions. Few studies, however, have systematically addressed the issue of topic development and loss in group interaction. In this paper, we are interested in answering a number of questions: Do men initiate topic changes more often when women have the floor? Do men and women initiate different types of topic transitions at different rates? Are sta- tus structures other than gender, especially those that develop in the course of the con- versation, important in structuring topic transitions?

Topics develop as speakers pursue, elabo- rate, and shape them. Topic loss occurs when a speaker's topic is ended with a topic transition. Presumably, determining what is talked about, which topics get developed and which are discarded, also has an impor- tant impact on a group's outputs. Speakers compete for the floor, not just in micro- level turn-taking, but in terms of the collectivity's attention to procedural strate- gies and potential solutions to group tasks. How these issues are resolved in a discus- sion often determines what the group does and who gets credit for its outputs.

To understand how topic transitions occur, it is useful to begin with the model of turn- taking designed by Sacks et al. (1974). This model specifies "rules" regarding appropri- ate turn-taking places in conversations. Ac- cording to this model, three options are available at the end of a completed word, phrase, or sentence: the current speaker may end his or her turn at talk by addressing a new speaker, another speaker may enter the conversation, or the current speaker may continue. When a speaker chooses one of the three options, he or she exhibits under- standing of prior talk. In other words, a con- versationalist must "fit" his or her current utterance to the utterance of the prior speaker (Schegloff, Jefferson, and Sacks 1977). A topic change occurs when an utter- ance does not show a clear sequential or ref- erential relationship to prior talk.

Maynard (1980) produced the most de- tailed analysis of the structure and function of topic transitions within conversation. He described a topic change as "an utterance directed to occasioning a new set of mentionables" (p. 280) and conceptualized topic changes as potential solutions to failed speaker transitions. When a current speaker stops and the other speaker does not immediately start up, Maynard referred to the resulting silence as the failure of a prior topic to yield successful transfer of speakership. A topic change restores smooth speaker transition, a solution to the problem.

Most studies examining topic change have focused on naturally occurring dyadic conversations, where both the topics and the structure of talk are open for negotiation between participants. Many of these studies have focused on gender inequalities. For example, in her study of conversation among intimate heterosexual couples, Fishman (1977, 1983) discovered that men and women raised similar numbers of top- ics, but men's attempts were successful 96 percent of the time, while women's attempts were successful only 36 percent of the time. Fishman (1977, 1983) did not dis- cuss the decision rules used to define a topic change, but her findings suggested a gendered status-ordering within conversations.

Many in the conversation analytic tradi- tion have criticized work from the 1970s and early 1980s that concentrated on gender and other social categories as isolated indepen- dent variables (Schegloff 1987:2 14-28; Thorne et al. 1983: 12-16). While this work (especially the research on interruptions by Tannen [1986], West 1979, West and Zimmerman [1983], and Zimmerman and West [1975]) attracted attention from other areas, the gender differences found often failed to hold up in varied contexts and in larger corpuses of conversation (see reviews by Aries 1996; James and Clarke 1988). As Thorne et al. (1983) point out, "The study of isolated social statuses like gender almost invariably leads to further questions about the effects of setting, topic, roles and other social factors that may interact with gender. It also raises questions about function and use" (p. 13). Schegloff (1987:215-16) argued that attention to simple individual at- tributes often fail to reflect the conversants' own orientation to the interaction and to de- flect attention away from the actual real- time mechanisms through which conversa- tional patterns are produced. We attempt to avoid these pitfalls in two ways. First, we use the conversation analytic literature pri- marily for insights about how to code topic transition structures in large corpuses of task-oriented conversation. Second, we highlight the dynamics of conversation that develop within the context of the group dis- cussions that we study, as well as examining the effects of gender.

West and Garcia (1988) were the first to in- troduce an explicit framework for analyzing topic transitions. The authors identified two types of topic transitions-collaborative and unilateral-that they found in mixed-sex, unacquainted dyads. A collaborative transition occurs when both participants in a con- versation jointly contribute to closing down a topic. A unilateral topic transition results from a noncollaborative topic shift on the part of one speaker. Unilateral topic transi- tions violate turn-taking norms by failing to acknowledge the conversational rights of the previous speaker: One participant exercises control over the topic, causing the other to experience topic loss. In their study of 12- minute conversations in five dyads, West and Garcia (1988) found that when topic changes were collaborative, women initiated as many topic changes as did men, but when unilateral topic changes occurred, men were usually the initiators. The authors used this evidence to argue that women are disadvan- taged relative to men in the allocation of "shift work," the procedures speakers use to produce shifts of topic.

Ainsworth-Vaughn (1992) expanded West and Garcia's coding scheme, conceptualizing unilateral, noncollaborative topic changes as part of a continuum in which transitions be- come less related to previous talk. She iden- tified two additional types of unilateral ac- tivities-links and minimal links-that varied in terms of how much acknowledgment was given to the previous speaker's contri- butions before moving on to the new topic. Ainsworth-Vaughn (1992) examined conver- sational dynamics in 12 physician-patient encounters and found that the ratios of col- laborative to unilateral activities differed widely for male and female physicians: For female physicians the ratio was 5 to 1; for male physicians the ratio was 1.4 to 1.

The conversation analytic studies reviewed above used an inductive approach to examine relatively small samples of talk and concentrated on the conceptual development of topic transition structures. We build on this research to formulate a systematic cod- ing framework for identifying topics which will allow us to study larger samples of talk. In addition, our analysis focuses on interac- tion in large task-oriented groups rather than in dyads in intimate relationships, naturally occurring conversations, or doctor-patient interviews. We do, however, analyze a small sample of dyads to test whether group size affects task-oriented topic transitions.

We expand the scope of topic transition research into group task-oriented interaction, which occurs in many institutional settings (e.g., juries, committees, work groups). Thus, we shift the focus of study from the original conversation analytic work where turn order, length of conversation, and con- tent were free to vary widely. We develop hypotheses about groups that are given a structure-a specific task, some limits on the length of time they have to work on the task, the types of solutions that are viable. Al- though these discussions are constrained, important transitions still occur in these dis- cussions-talk that shifts the group's atten- tion from one part of the task to another, comments that call attention to procedures the group should adopt, and (perhaps most important) proposed solutions to the group task. We believe that many of the features that have been discovered in more free-rang- ing conversations are relevant to these group task discussions as well. Therefore, we hope to broaden the application of these ideas to this somewhat different, more structured do- main. If this generalization proves fruitful, it will be important, as significant work gets done in settings that are structured by some external institutional framework.

In particular, we are interested in whether some of the inequalities that conversation analysts have noted (e.g., with regard to gen- der and other social statuses) structure the content of transitions within task-oriented groups. If so, these topic transitions would constitute a part of the power and prestige order of the group, the way in which it orga- nizes its activity to produce a high-quality group solution to a task.

Expectation states theory, with its roots in exchange theory and in Bales's classic small group research (Bales 1951), suggests that people in a collaborative, task-oriented group will form expectations about the rela- tive competence of group members to make useful contributions to the task. When per- formance expectations for a group member are high, he or she will (1) be given more opportunities to participate, (2) offer more task contributions to the group, (3) receive more positive evaluations for his or her task contributions, and (4) have more influence over other group members (Berger, Conner, and Fisek 1974; Berger et al. 1977). These interrelated behaviors, all of which direct the group's task activity toward quality solu- tions, create the power and prestige structure of the group. We argue that topic transitions within a structured task-oriented group dis- cussion contribute to this power and prestige order.


Engaging in status-related behaviors during task performance tends to stabilize the struc- ture of conversational inequality among group members. We consider topic transi- tions to be status-related behaviors because they represent opportunities to participate in the group's work. For example, a collabora- tive transition occurs when group members jointly contribute to close a topic down. This type of transition offers an opportunity for someone to speak up, avoid a failed speaker transfer, and direct the group to a new topic. Clearly, this constitutes an action opportu- nity in the expectation states framework.

Other types of topic transitions also may act as status-related behaviors because they can be used to evaluate and influence oth- ers. Initiating a linked topic transition is one way a group member can negatively evalu- ate another by cutting off a topic that is not useful for the collective goal. A group mem- ber can also use a suddenlunilateral transi- tion to influence the group, as this type of transition secures control over the proce- dures and ideas that will contribute to the task solution.

If topic transitions help create the power and prestige order, then the theoretical state- ments of expectation states theory should apply to them. We therefore summarize pre- dictions from two important branches of the theoretical research program: (1) status char- acteristics theory, and (2) new work on en- dogenous status dynamics.

The status characteristics branch of expecta- tion states theory describes how information from the larger social structure is imported into groups to form expectations. If people generally expect one category of people to be more competent than another at most tasks, this distinction is called a status char- acteristic. Unless specific information about the task contradicts the general expectation state attached to a status characteristic, group members will treat the characteristic as if it were relevant to the task at hand and it will influence power and prestige behaviors (Berger, Cohen, and Zelditch 1972; Berger et al. 1977). Studies have demonstrated that gender, once activated as a salient character- istic in a mixed-sex group, shapes perfor- mance expectations that influence conversa- tional behavior (Balkwell and Berger 1996; Dovidio et al. 1988; Ridgeway, Berger, and Smith 1985; Wood and Karten 1986).* If ini- tiating topics is a status-related task behav- ior, then men should be more likely than women to take these action opportunities. Thus, in a speaker transition between the cur- rent speaker (who has the potential of losing a topic) and a following speaker (who has the potential of initiating a new topic, thereby cutting short the previous topic), Hypothesis 1 predicts a gender-of-follower effect. Given that the gender of follower variable is coded male = 1, female = 0, a positive effect would suggest that men are more likely to initiate transitions than women.

Hypothesis 1:In mixed-sex groups, men will

initiate more topic transitions than will


When women do make contributions to the group, their ideas are more likely to be evaluated negatively than are men's. When a group member receives a topic transition, his or her ideas are not developed by others, but instead are cut off or die from inatten- tion. Because this phenomenon of topic loss represents an implicit (in collaborative tran- sitions) or explicit (in more unilateral topic transitions) negative evaluation of the ideas currently on the floor, we expect women to

A legitimacy process may develop in which group members socially construct a normative "right" to earn status for those associated with the high state of the status characteristic. Active and assertive contributions on the part of low-sta- tus members may be seen as illegitimate and sub- ject to others' sanctions. Past research supports the idea that low-status members who attempt to gain status in the group by actively contributing to the group task often draw a negative reaction from other group members (Ellyson, Dovidio, and Brown 1992; Meeker and Weitzel-O'Neill 1977; Ridgeway 1982; Ridgeway and Berger 1986).

be the recipients of topic transitions more often than are men. Hypothesis 2 predicts a negative gender-of-speaker effect in a tran- sition between an initial speaker and a fol- lowing speaker.

Hypothesis 2: In mixed-sex groups, women will receive more topic transitions than will men.

Since men typically have higher social sta- tus than women, it is more appropriate for men than for women to evaluate other group members negatively. If a male group mem- ber initiates a topic change, his behavior may be seen as warranted because he has high status. Legitimacy processes reinforce the predicted pattern. High expectation members of the group (in this case, males) have greater rights to discriminate among other group members in handing out inter- actional rewards and sanctions. Thus, Hy- pothesis 3 predicts a negative gender-of- speaker x gender-of-follower effect. This in- teraction suggests that the gender-of-fol- lower effect varies depending on whether the gender of speaker is male or female; in this case, when men are the following speakers who have the potential of cutting off the topic of the current speaker, they are more likely to do so when women are the current speakers.

Hypothesis 3:In mixed-sex groups, men will direct more transitions at women than at men.

Conversation analysts identify four types of topic transitions arrayed in order of their individual versus collective origin-sudden/ unilateral, minimally linked, linked, and col- laborative. Of these four types, the sudden transition effectively prevents the current speaker from completing his or her thoughts on the current topic and introduces new, "better" ideas. We expect male group mem- bers to initiate sudden topic transitions more often than females for two reasons. A sudden topic transition is both a task contribu- tion and a clear negative comment on the current speaker's contribution. Because high-status group members are expected to make more contributions and to judge the contributions of others, we expect men to engage in this behavior more often. On one hand, females are not expected to use sud- den transitions as often as men because their lower status would make such a judgmental intrusion seem illegitimate. On the other hand, because women are associated with lower performance expectations, their ideas are more likely to be evaluated negatively. Thus, they will receive more sudden transi- tions from group members. Men, however, will be less likely to receive sudden transi- tions.

Other forms of transitions-minimally linked, linked, and collaborative-not only show increasingly more connection to and acknowledgment of previous talk, but these transitions also demonstrate progressively less negative evaluation of the current speaker's ideas. They still represent task contributions and therefore may more com- monly come from high-status group mem- bers. But when women wish to move the group forward in a productive way, they will be more likely to use collaborative structures than intrusive ones because collaborative ac- tions will seem more group-oriented and therefore be perceived by others as legiti- mate. Therefore, men and women will be relatively more equal in their use of linked or collaborative topic transitions. These types of transitions represent the acceptance of an action opportunity, but they do not in- volve a negative comment on another group member's topic development. Indeed, women may engage in a higher rate of col- laborative transitions than men, as these transitions may represent task contributions that show group motivation and are therefore legitimate for low-status members (Ridgeway 1982). Thus, we predict:

Hypothesis 4: The effects in Hypothesis 1 through Hypothesis 3 will be strongest for sudden transitions, intermediate for linked topic transitions, and least strong or even reversed for collaborative topic transitions.


The predictions for group interaction in ex- pectation states theory depend on each group member comparing their expectations regarding their own competence at the task with their expectations for each of their fel- low group members (Skvoretz 1981; Smith- Lovin, Skvoretz, and Hudson 1986). There- fore, group composition is predicted to af- fect group interaction patterns only through the salience process (i.e., characteristics that differentiate group members are more likely to become salient in competence evalua- tions) and through the dyadic comparisons that are embedded in the group (e.g., a fe- male who looks out at a group and sees her- self as inferior to all five other male group members will take fewer action opportuni- ties than one who is status-equal to four women in the group, but status inferior to one male). The dyadic comparison effects are captured by Hypotheses 1 through 3.

Cultural approaches to explaining gender differences, however, argue that the effects of group composition are stronger than the simple sum of the dyadic comparisons em- bedded within the group structure. Cultural researchers argue that girls and boys learn how to use language in different ways through their play activities in same-sex groups (Maltz and Borker 1982; Tannen 1986, 1990a, 1990b). Boys play in groups with a competitive structure, in which they learn to maintain an audience and assert themselves; girls create smaller groups of tight friendships by exchanging information and confidences (Goodwin 1980; Maltz and Borker 1982). School-age children learn gender-specific "cultures" through interac- tions that carry over into their adult lives.

This perspective emphasizes the different cultures that are evoked in groups that are dominated by one gender. Women's talk is seen as more friendly, cooperative, and re- lational (Carli 1989; Fishman 1983; Lakoff 1973; Ridgeway and Johnson 1990). Men's talk is more dominant, more directive, and less supportive. For example, in male groups, members are more likely to chal- lenge or dispute another speaker's utter- ances and to offer no response or acknowl- edgment to the comments of another speaker (Fishman 1983; Zimmerman and West 1975). This leads us to predict that:

Hypothesis 5:Groups containing proportion- ately more women will produce a higher rate of collaborative topic transitions relative to other types of topic transi- tions when compared with groups with proportionately more men.

Hypothesis 6:Groups containing proportion- ately more men will produce a higher rate of sudden topic transitions relative to other types of topic transitions when compared with groups with proportion- ately more women.


The arguments above focus on how the char- acteristics that people bring with them to the group (e.g., gender) influence the group's conversational structures. However, dy- namic processes that develop within a con- versation can powerfully shape the distribu- tion of these conversational behaviors. An- other tradition within expectation states theory formally develops ideas about these processes: expectation states as dynamic properties.

Expectation states theory would predict a gendered pattern of topic transitions only if gender were a salient status characteristic in these task groups. Although gender typically operates this way in our culture, it is a rela- tively weak status characteristic (although one that can transform itself through a vari- ety of evolutionary changes in structural gender relations) (Ridgeway and Smith- Lovin 1999). Among otherwise status-equal college undergraduates, gender may be a sta- tus-organizing characteristic that can be eas- ily overcome with other information about expectations for the task at hand (Riordan 1983; Wagner and Berger 1997). This be- comes an important issue when we consider the dynamics through which status structures are generated (Fisek, Berger, and Norman 199 1; Skvoretz and Fararo 1996).

Expectation states research recognizes that participation hierarchies can form in initially status-homogeneous groups. In these groups, actors are linked to expectation states through a process of differentiation based on their behaviors during the group's discussion (Berger et al. 1974). Recent work elaborates how status characteristics combine with be- haviors that are displayed during a task-ori- ented group discussion to create an expecta- tion-ordered conversational structure (Fisek et al. 1991; Skvoretz and Fararo 1996). In a sense, the status characteristics and evalua- tion expectation branches of the theoretical research program combine in the study of any particular task group. Skvoretz and Fararo (1996) emphasize the probabilistic nature of the operation of status characteris- tics within groups and their model makes clear that a relatively weak status character- istic can easily be overcome by task cues that are displayed in the context of group discussions.

Similarly, in his study of discussions in naturally occurring managerial groups, Gibson (1998) finds that differences in man- agers' volubility (self-selection into speak- ing) often block the effects of status that are external to the group. Thus, a high-status person must first speak to be addressed by other group members with high frequency. Although turn-receiving (the tendency to speak after being addressed) is the primary mechanism through which high-status people achieve centrality in group discus- sions, they often must prime that process by offering contributions. A high-status person who does not make such contributions early in a group discussion is often superceded by other group members who are frequently ad- dressed because of earlier contributions that they have made.

In our data, we can explore these dynam- ics by using early participation in the group discussion as an indicator of an internally generated status hierarchy. Here we rely on the well-established finding that participa- tion is both a task cue indicating high expec- tations and an outcome of subtle, unmea- sured status cues about which we have no data (e.g., social class displayed through dress, demeanor, or personal attractiveness). We avoid the problem that participation is both a cue and an outcome of status-orga- nizing processes by using participation prior to the topic transition being predicted. We therefore ask: How does prior participation in a group discussion influence task transi- tion behavior? We propose hypotheses that directly parallel our hypotheses about gen- der as a status characteristic, only here prior participation in the group discussion acts as a task cue indicating expectations for com- petence at the task.

Hypothesis 7 predicts a positive participa- tion-of-follower effect, which would suggest that the higher levels of participation by a speaker will be associated with the initiation of more topic transitions. By using the par- ticipation of the following speaker in the analysis, we are able to find out if high par- ticipation rates and high topic initiation rates move hand in hand.

Hypothesis 7: Speakers who have contrib- uted more to the group discussion will initiate more topic transitions than will low-participation speakers.

Hypothesis 8 predicts a negative partici- pation-of-speaker effect; this predicted ef- fect suggests that low levels of participation on the part of the speaker will be associated with higher levels of topic loss. Using the participation-of-speaker variable, we can find out whether low levels of participation in the group discussion do in fact lead to more topic loss.

Hypothesis 8: Speakers who have contrib- uted less to the group discussion will be more likely to lose topics that they are developing than high-participation speakers.

Hypothesis 9 predicts a participation-of- speaker x participation-of-follower effect. Such an interaction suggests that the partici- pation-of-follower effect varies depending on whether the participation of the speaker is high or low; in this case, high-participation speakers will initiate topic transitions when low-participation speakers have the floor.

Hypothesis 9: High participation speakers will direct more topic transitions toward low-participation speakers than toward other high-participation speakers.

Finally, to construct hypotheses parallel to the gender hypotheses, we predict the fol- lowing:

Hypothesis 10: The effects in Hypothesis 7 through Hypothesis 9 will be strongest for sudden transitions, intermediate for linked topic transitions, and least strong or even reversed for collaborative topic transitions.

In addition, we argue that if topic transi- tions are status-related behaviors and act as task cues to form expectation states, topic transitions will tend to be stable over the course of a group conversation. Those who

lose topics early in the discussion should be more at risk of topic loss later. Similarly, those who initiate new topics early in the conversation are giving task cues that will lead others to define them as occupying a high expectation state. Their topics are more likely to be developed (and less likely to be lost) later in the discussion. In effect, we predict that topic initiation and loss consti- tute somewhat stable conversational roles, as group members develop expectations for one another and act on the basis of those expec- tations.

Hypothesis 11: Topic initiation early in the group discussion will reduce the likeli- hood of topic initiation late in the dis- cussion.

Hypothesis 12:Topic loss early in the group discussion will increase the risk of topic loss later in the discussion.


We use two data sets in this study. The first

data set includes conversations in 22 six-per

son, task-oriented, mixed-sex groups (Smith-

Lovin et al. 1986). We use these groups to

test the 12 hypotheses developed above. We

then replicate the analyses on a sample of

dyadic task interactions collected especially

for this study. By using the same task and

experimental procedures, the dyad data will

allow us to see whether the size of the group

in our first analysis is responsible for the pat-

terns we find.

Subjects in the group study were under-

graduates at the University of South Caro-

lina who enrolled in an introductory sociol-

ogy class in the early 1980s. The experiment-

ers formed groups of six previously unac-

quainted undergraduates in which gender

composition systematically varied. Partici-

pants were told that they were part of a group

problem solving study and were directed to

work collectively to reach a group decision.

Each group was given 30 minutes to solve a

sex-neutral task."he task was to formulate

The task was developed by Fisek (1974) and met the scope conditions for expectation states theory.

a problem for another group to solve that fit the following criteria: (1) group members could not have any special knowledge or fa- miliarity with the problem area, (2) the prob- lem had to be one that would force group members to use their own judgments in solv- ing it, (3) the problem could not involve any predetermined and strongly held values or opinions, and (4) the problem had to be of interest to group members. The group was required to write down the problem that the group chose, how it fit the specified criteria, and the recommendations for presentation of the problem to another group. Groups were given a collective incentive to perform well-$30 was awarded to the group with the best task ~olution.~

We use 291 pages of tran- scribed discussion from the 22 mixed-sex groups in our analysis (Smith-Lovin 1990).

Building upon the conversation analytic re- search, we developed a systematic coding scheme for topic transitions in the task-ori- ented groups. First, we identified when the groups moved from one topic to another in the course of their discussion using features highlighted by Maynard (1980). Maynard defined topicality as "partly constituted in the procedures conversationalists utilize to display understanding and to achieve one turn's proper fit with a prior" (p. 263, italics in original). He went on to describe it as "an achievement of conversationalists, something organized and made observable in pat- terned ways that can be described." In other words, understanding can be demonstrated through the use of acknowledgments, substi- tutions, and deletions-all conversational mechanisms that facilitate the achievement of topicality. For example, acknowledgment cues such as "mm-hmm" or "yeah" demon- strate attentiveness and comprehension, and often encourage the speaker to continue in the same line of talk:5

See Smith-Lovin et al. (1986) for a more complete discussion of data collection proce- dures.

The following examples are taken from the transcripts compiled by Smith-Lovin (1990). Ital- ics emphasize the conversational features under discussion. Topic changes are marked by IT/. Other transcription features are adopted from
(Group 12; p. 116)

3: I think it would be interesting if jobs they have in school were in some way related to what they want to do in the future.

6: Yeah.

5: Hmm

Another way a speaker can demonstrate understanding of prior talk is through sub- stitution practices. For example, a speaker can use words such as "it" and "they" to re- fer to what the previous speaker had spoken about in the previous turn (Goffman 1983; Sacks 1967):

(Group 2; p. 18)

4: What about the judicial system, 1 like criminals, yeah. You know, how, how victims are just overlooked and every- thing.

11: Yeah.

1: Well, somebody would know about it, though. Somebody would have a special interest in it.

Finally, a speaker can display understanding through deletions. When deletions are used in a turn, a speaker's utterance makes sense only in the context of the prior talk (Goffman 1983):

(Group 18; p. 162)

1: OK, say we were going to say something about acceptance -All right, you said something about acceptance, what about it?

conversation analysts, although the detail pro- vided here does not approach the more refined analyses that researchers in that tradition require. The transcriptions use the following conventions:

-Pause within a sentence of at least .5 seconds, a break caused by an interruption, or a trail- ing pause at the end of sentence.

[ Before the number of the speaker, this refers to an interruption by that speaker, causing the initial speaker to discontinue talking.

/ (1) Within a turn, this refers to the point at which another speaker interrupts or overlaps the speech of the current speaker.

/ (2) Before the number of the speaker, this re- fers to simultaneous talk in the form of an in- terruption or overlap, in which the speaker is talking at the same time as the previous speaker.

6: I said -you know, being turned down. 1 You know, you build up your nerve to go ask her out and she turns you down. Makes you feel kind of I small. I know.

11: Rejection.

12: Shattered.

1: How can we work

[2: Then you're afraid to ask her out again.

Thus, speakers display understanding of the prior turn at talk in the form of acknowl- edgment cues, deletions, or substitutions that indicate the continuing development of a topic. The absence of these features indi- cates transition to a new topic (indicated in the following excerpts by "IT/"). After we identified topic transitions in the group dis- cussion, we further classified each transition as collaborative, linked, minimally linked or sudden, using Ainsworth-Vaughn's (1992) coding criteria.

TOPIC TRANSITIONS. A collaborative transition was coded when speakers cooperatively "closed down a topic" through the exchange of objects such as "well," "okay," and "all right" (West and Garcia 1988):

(Group 12: p. 117)

5: You'd be in kind of a little trouble there 1 if you applied value or opinion.

14: Really.

1: It is of interest to most of us.

6: Definitely.

2: Mm-hmm.

6: OK.

2: /TI We've got five minutes till it's time out.

A collaborative topic transition was also coded when participants jointly engaged in summarizing activities to bring the topic on the floor to a close, thereby opening the pos- sibility for a topic change. In this type of re- ciprocal activity, one participant summarizes or assesses the topic on the floor and another affirms the assessment (Ainsworth-Vaughn 1992):

(Group 12; p. 112)

4: How about finances in college?

2: That's rather controversial.

4: That's a strongly held opinion.

2: Yeah, that's strongly held.

5: IT/ How can we avoid another tuition increase? What could be, uh, what type of program in school could be elimi- nated?

The topic on the floor can also become "extinct" or "die" through lack of verbal ac- tivity (West and Garcia 1988). Referring back to the turn-taking structure of conver- sation, silences can occur during turn-trans- fers. Maynard (1980) called this silence the failure of a prior topic to yield successful transfer of speakership. Thus, a topic change is a solution to this problem:

(Group 31; p. 268)

3: Money -"problem must be of interest to most college students" -money.

6: But everybody has strong opinions about it.

3: That's true.

2: Are they that strong?

3: If you I want money.

16: If you don't have money.


3: Um.


5: /TI How about something with disabili- ties, have something about blindness or something like that, some handicap.

A dummy variable for collaborative tran- sitions was coded 1 when any of the above situations occurred among group members, and coded 0 if another type of transition oc- curred.

UNILATERAL TOPIC TRANSITIONS. Three types of unilateral topic transitions were identified: linked, minimal link, and sudden. When initiated, all three prevent the continu- ation of the current topic. A linked transition refers back to the prior turn at talk before in- troducing a new area of discussion. This type of transition does not contribute to topic development and can be recognized by the reference to previous talk before changing the topic:

(Group 20; p. 170)

5: The economy, something like that.

1: Hmm?

5: Economy.

1: Or economy -/TI Then again, with mar- riage, though, it would probably get in a fight over that because some people

Here Speaker 1 repeats Speaker 5's sugges- tion of the economy as part of the group so- lution, clearly acknowledging the task con- tribution before changing the topic to mar- riage. This linked transition is considered unilateral because Speaker 5 may not have finished her thoughts about how the economy could be an effective solution.

A minimal link transition refers back to the previous turn at talk using only minimal re- sponses or acknowledgment cues, such as "all right" or "mm-hmm." In this instance, these minimal responses are not considered back-channel cues (interjections that support and facilitate the speaker's talk) because they do not encourage the speaker to elabo- rate on a topic:

(Group 3; p. 33)

6: Start writing. Who's got the prettiest handwriting? I know it's not Skip and me.

1: Definitely not him.

4: /T/ Okay, "Brief statement of the prob- lem." How do you want to state it, say, "The park on the corner of Bull and Blossom"

In this example, Speaker 4 acknowledges the previous talk about handwriting by saying "okay," but then attempts to move the dis- cussion forward by referring to another as- pect of the task. This type of transition is considered unilateral because even though Speaker 4 refers to prior talk, there is no col- laborative closure.

A sudden transition results from a unilat- eral shift on the part of a participant to talk. To engage in a sudden topic change is to ex- ercise control over the topic. In this situa- tion, control is not shared, and a participant's topic may be lost (Ainsworth-Vaughn 1992). We identify a sudden transition when a speaker changes the topic without expressing understanding of the previous turn at talk. A speaker who does not use acknowl- edgment cues, deletions, or substitutions in a turn is disregarding the previous utterance, thereby initiating a sudden topic change:

(Group 2; p. 18)

5: Why don't we look at national prob- lems. Everyone has a problem with President Reagan. Everybody has differ- ent views

4: I like him [defensively].

5: I like him, too. But a lot of people think I he's making, you know, the world so much worse. I happen to think he's do- ing a good job. That's my opinion, ev- erybody has his own opinions to him.

13: -he's the worst president.

3: Once again, / I think that goes against #3, stay away from political

15: What he's done so far

2: IT/ What about something like the Co- lumbia? I mean, it's just gone up in space a third time.

Here the discussion about President Reagan as a candidate for the solution to the group's task has not been completed when Speaker 2 introduces the topic of the Columbia space shuttle. Even though Speaker 3 begins to disagree with the suggestion of President Reagan and national problems, Speaker 5's opportunity to defend his idea is cut short because of Speaker 2's sudden topic change. One could speculate that Speaker 2 may be responding to Speaker 5's initial statement about national problems. However, Speaker 2 does not bring up a national problem, but a national issue, and her comments have no reference to the previous turn at talk.6

Another example of a unilateral topic change shows how a shift from substantive task solutions to procedural matters can de- rail a topic's development:

(Group 12; 113)

3: A problem, think of a problem.

2: You can't do anything about how to make sure a student makes at least a C in a sociology course. If you don't like a professor's performance, you can't do anything about that, or course structures, or

5: IT/ "Recommendations on how the problem should be presented to the group."

Here Speaker 2 begins to introduce a prob- lem, a potential solution for the group task. He does not finish his thought because

One might argue that Speaker 2 is introduc- ing a new topic to move the discussion away from a divisive, conflictual topic. Because this group-oriented type of topic transition may be more common in large, task-oriented discussion groups, we use the dyad study to examine the ef- fects of group size.

Speaker 5 unilaterally changes the topic. Af- ter Speaker 5's contribution turns the group's attention to procedural matters, Speaker 2's comments are never revisited.

Two researchers independently coded the topic transitions in each of the 22 groups. The gender of each speaker within each group was not known to the coders. Intercoder reliability (Cronbach's alpha) was .86.' Disagreements between the two coders about transitions were resolved by having the transitions blind-coded by a third researcher and then discussing the differing assessment with all three coders.

We created a dummy variable for each type of topic transition. For example, the dummy variable for sudden transitions was coded 1 when a sudden transition occurred, and coded 0 when the topic ended through another type of transition.


We replicated the original experiment using dyads to explore whether the six-person groups produced different conversational structures from the dyads. The subjects were undergraduates enrolled in an introductory sociology class at a large southwestern uni- versity in 1996. Twenty unacquainted dyads, each including one male and one female, par- ticipated in the experiment. Each experimen- tal session was videotaped and transcribed. The analysis here uses data for 19 dyads8 The reliability of topic transition coding for the dyad study is Cronbach's alpha =

We use a continuous-time event history model (Allison 1984; Yamaguchi 1991) to assess the impact of gender, group composi-

The intercoder reliability is higher (Cronbach's alpha = .89) for the simpler judg- ment of whether or not a topic transition occurred than for the judgment of what type of transition occurred. The four types of topic transition did not differ significantly in their reliability.

We excluded one session because the dyad was mixed by race, which introduced another po- tential status characteristic that might confound the analysis.

Here, we coded only whether or not a topic transition occurred, not the type of transition.

tion, and participation on the rate of topic transitions. This method is appropriate given that we are interested in a dynamic approach to topic transitions that allows for the use of information about the timing of events and changes in the levels of covariates. This model provides estimates of the effects of covariates on an unobserved rate of occur- rence for some event. We use the propor- tional hazards model in following form,

where a(t) is an unspecified function of time. Because this function does not have to be specified, the model is often described as semiparametric. The hazard rate, h(t), is the instantaneous risk of the event occurring at time t, given that the event did not occur be- fore time t. The dependent variable in this analysis is the hazard rate of a topic transi- tion occurring at the end of an utterance.

The unit of analysis is the utterance, or turn at talk. There are 4,184 transitions or turns in the 22 six-person group conversa- tions. Of these events, 3,336 are censored (normal, nontopic-changing transitions), 848 are uncensored and represent the occurrence of topic transitions. Of the 848 topic transi- tions, 235 are collaborative, 97 are mini- mally linked, 139 are linked, and 377 are sudden transitions.

In the dyad replication study, there are 3,707 transitions or turns in the 19 conver- sations; 3,393 of these events are censored, representing normal, non-topic changing transitions, and 314 are uncensored, repre- senting the occurrence of a topic transition at the end of an utterance. We do not subdi- vide the topic transitions for dyads into dif- ferent types (collaborative, linked and sud- den) as this would create too small a sample size for meaningful event history analysis.


GENDEROF SPEAKER. The gender of the current speaker (who has the potential of losing a topic) is coded 1 = male, 0 = female.

GENDER OF FOLLOWER. The gender of the speaker who follows the current speaker (who has the potential of initiating a new


topic) is coded 1 = male, 0 = female.

GROUPCOMPOSITION. A variable repre- senting the number of females in each group, which ranges from 1 to 5, is included in the analysis. Because all groups contained six participants, this variable also represents the inverse of the number of males in the group. (In the dyadic replication, all dyads were male-female.)

PARTICIPATION OF SPEAKER AND FOL- LOWER. TO examine whether the conversa- tions are status-organized in ways other than gender, we include the following variables: participation of speaker (the number of words the current speaker has spoken so far in the discussion), participation of follower (the number of words the next speaker has spoken so far in the discussion), and the multiplicative interaction of those two vari- ables.

TOPICHISTORY. We count the number of times that the current speaker has experi- enced topic loss in the group discussion (pre- vious topic loss by speaker) and the number of times that the follower has introduced new topics (previous topic initiation by follower). These variables are time-varying, as they start at zero for all participants and are incremented by one each time a group mem- ber either has the topic that they are develop- ing changed, or introduces a new topic.

PAUSES.Maynard (1980) argues that ex- ceptionally long pauses indicate that a topic is failing to sustain conversation. We note whether a speaker transition is preceded by a pause of longer than .5 seconds to assess whether this marker of conversational dis- continuity affects the risk of topic loss. The pause variable is coded 1 if a pause precedes the transition, 0 otherwise.


Maynard (1980) also notes that very short utterances indicate a lack of interest in the current topic and its failure to sustain nor- mal turn-taking. Here we code the length of the utterance (in number of words) before the speaker transition to see if short utter- ances are more likely to precede topic loss.

SPEAKERRIGHTS. We thought it Was likely that the person who had introduced a topic might have special rights to end its de- velopment and to move the group to another, more fruitful, line of discussion. We there- fore include a dummy variable coded 1 if the person who initiated a topic is the next speaker (in other words, is also the potential topic changer), 0 otherwise.

TOPICDURATION SINCE LAST TRANSI- TION.We measure the duration since the last topic transition as the number of words since the last topic transition occurred, to check for time dependence within topics.

LOG OF TIME SINCE BEGINNING OF CON- VERSATION. The data exhibit evidence of time dependence over the course of the dis- cussion, which is controlled by including the log of time (in number of words) since the beginning of the conversation.


Our preliminary evidence suggests that topic transitions are related to the status and influ- ence processes within these collectively ori- ented task groups. First, topic initiation is closely related to participation in the groups, a variable that researchers typically assume is related to task contribution and influence over the group's task solutions. The top par- ticipator (measured by number of words spo- ken in the discussion) in each of our 22 groups interjected an average of 316 topics over the course of the group discussions, while the lowest participator suggested just 28 topics. This ratio of 11.1 to 1 is a more striking inequality than we see in participa- tion, where each top participator took an av- erage 1,327 turns at talk in the groups while the lowest participator took just 27 1 (a ratio of 4.9 to 1). More direct evidence of the re- lationship between topic transitions and in- fluence is the fact that the group member who suggested the task solution that the group finally adopted, initiated an average of

10.5 topics during the group discussion. The other five group members together initiated an average of only 5.9 topics. Of course, it is still possible that topic initiators are suc- cessful at influencing the group discussion only by chance (e.g., through the fact that they proposed more ideas and that one of their ideas was more likely to be chosen at random). The strongest evidence against this baseline argument is a status structuring of topic initiation and loss. We now move to considering these patterns.

Table 1. Frequency Distribution of Topic Transitions by Gender of Follower and Gender of Speaker in Mixed-Sex Task Discussions: Six-Person Groups and Dyads
Gender of Follower     Gender of Speaker
    (Initiator)            (Receiver)     
Tvue of Grouo Male         Female     Male         Female

Six-person groups a 460 388 477 371

(54.2) (45.8) (56.3) (43.7)

Dyads 166 148 149 165

(50.1) (49.9) (50.1) (49.9)

Note: Numbers in parentheses are percentages. a Twenty-two six-person task-oriented mixed-sex groups of undergraduate sociology students from the University of South Carolina (Smith-Lovin, Skvoretz, and Hudson 1986); N = 132. Numbers of females and males in each group varied. Nineteen male-female dyads from an introductory sociology class at a large southwestern university (data collected by the authors, 1996); N = 38.

Table 2. Frequency Distribution of Different Types of Topic Transitions by Gender of Follower and Gender of Speaker in Mixed-Sex Task Discussions: Six-Person Groups
Type of Topic Transition     Gender of Follower (Initiator) Male Female     Gender of Speaker (Receiver) Male Female
Collaborative     133 (28.9)         102 (26.3)     133 (27.9)         102 (27.5)

Minimal link


Total 460 388 477 37 1

(100.0) (100.0) (100.0) (100.0)

Note: Numbers in parentheses are percentages. See Table 1 for description of data source.

GENDERED? and women are even more similar in initiat-

ing and receiving topic transitions, a func- Table 1 indicates that topic transitions are tion of the fact that the participants in dy- similarly distributed among men and women adic discussions tended to alternate topics in mixed-sex task discussions. While men throughout the conver~ation.'~ initiate somewhat more new topics than do The minimal effect of gender is reinforced women in the six-person groups (460 com-by the close correspondence between pared to 388, or a ratio of 1.1 to I), they also women's and men's types of transitions lose more topics that they are developing

(1.3 to 1). Both of these differences seem to

lo Alternating topics does not imply that there

be driven more by the fact that men talk

are no gender differences in topic development

more than women in these groups (Smith-

and loss in the dyads, however. It may still be Lovin and Brody 1989; Smith-Lovin et al. the case that women's topics are developed for 1986) than by any gender effect on topic shorter periods of time and are lost more quickly transition dynamics. In the dyadic data, men than are men's topics.

(Table 2). The largest gender differences in Table 2 are the 2.6 percentage-point differ- ence between male and female initiation of collaborative transitions (with the males showing a higher percentage of collabora- tive topic shifts) and the 2.6 percentage- point difference between male and female topic loss through linked transitions. The overall impression is one of substantial simi- larity between male and female patterns of topic transition, both as initiators of topic change and as speakers experiencing topic loss. However, these simple percentages fail to account for the amount of time men and women spend contributing to the group dis- cussion, which determines their exposure to the risk of topic change. In addition, the de- scriptive data cannot assess whether men's or women's topics are more successfully de- veloped (i.e., survive for longer periods of discussion). Event history analyses provide more sophisticated modeling of the dynam- ics of topic development and loss in these groups.

Column 1 in Table 3 shows the hypothesized direction of effects for each variable and col- umns 2 and 3 show the results of our event history analysis of all topic transitions in the six-person and dyadic task-oriented discus- sions.

GENDER.The results for the six-person groups (column 2) indicate that there are no main effects of gender on the hazard rate of topic transitions in the group conversations." Men are not more likely to initiate new topics for discussion (gender-of-fol- lower effect), so Hypothesis 1 is not sup- ported. Women are not more likely to expe- rience the loss of topics that they are devel- oping (gender-of-speaker effect), so Hypoth-

l' In this analysis, we are interested in the gen- der of the speaker whose topic is changed by the following speaker, and the gender of the follower who changes the topic. We do not include the gender of the speaker who introduced a specific topic. In analyses not shown here, however, we added gender of topic originator in all of the models. We found that including the variable did not change the original results, nor was the coef- ficient for the variable itself significant.

Table 3. The Effects of Gender, Participation, and Conversation Analysis Variables on the Hazard Rate of Topic Transitions: Six-Person Groups and Dyads

(1) (2) (3) Independent Six-Person Variable Prediction Groups Dyads

Gender Variables

Gender of speaker --.031 ,782 (.107) (.788)

Gender of follower + -.05 1 .9 18 (.108) (.788)

Gender of speaker x --.608*** NA Gender of follower (.149)

Group composition -.134*** NA (no. of females) (.032)

Participation Variables

Participation of -

Participation of +

Participation of -
speaker x
of follower

Previous topic loss +
by speaker

Previous topic +
initiation by

Topic Change Marker Variables

Pause + .267* .893* (.156) (.506)

Number of words --.009*** -.006 in previous turn (.003) (.004)

Speaker rights + 1.13*** N A (.09 1)

Control Variables

Topic duration since -.003*** -.001*** last transition (.001) (.OOO)

Time since beginning -.198*** ,105 of conversation (log) (.063) (. 143)

Number of cases 848 314 Degrees of freedom 14 9 -2 log-likelihood 10,677.7 4,597.0 Chi-square 2,387.87 62.72

Note: Numbers in parentheses are estimated stan- dard errors. See Table 1 for description of data source.

*p< .05 **p< .Ol ***p< ,001 (two-tailed tests)

esis 2 is not supported. There is, however, a negative effect for the interaction between the two gender variables. This interaction indicates that when a man introduces a new topic, he is more likely than a woman to dis- criminate in his evaluations of men's and women's topic development. This interac- tion is consistent with our argument in Hy- pothesis 3: Men, as high-status actors, can more legitimately evaluate the contributions of others and, in particular, can more readily dismiss the contributions of women. There is a significant negative effect of the group's sex composition, with more heavily female groups tending to develop topics longer be- fore shifting them.

Because our results differ from what we had predicted, with no main effects of gen- der, we explore these processes again in the dyad replication study. This also allows us to explore the possibility that topic transi- tions operate differently in dyads. Column 3 presents parallel analyses for the dyadic data and here, too, the gender main effects are not significant.l2 Therefore, we conclude that topic transitions in these task-oriented dis- cussions are not strongly gendered,13 al- though there is a greater tendency for men to discriminate between male and female speakers when they change a topic.

ENDOGENOUSCONVERSATION PATTERNS. We now turn to the time-varying variables that indicate status patterns that develop within the context of the conversa- tions. We find participation effects that indi- cate a status-organizing process in topic transitions. In the six-person groups, high- participation speakers are less likely to lose

l2 The gender interaction is not applicable to the dyadic discussions, as people almost never changed the subject while they themselves were developing a topic; therefore, the follower is al- most always the other sex from the speaker.

l3 One might expect that the effects of gender might be suppressed by the fact that participation is included in the model because participation and gender are somewhat related in these data (Smith-Lovin et al. 1986). However, controlling for participation creates only minor changes in the coefficients for gender. There are no substan- tial main effects of gender on topic transitions, even when the gender variables are the only pre- dictors in an event history model (results are available from authors).

a topic in a topic transition (negative partici- pation-of-speaker effect); Hypothesis 7 is supported. In other words, low-participation (presumably low expectation) speakers are at higher risk of topic loss. High-participa- tion speakers, however, do not appear to in- troduce topics at a higher rate. Since there is no participation-of-follower effect, Hypoth- esis 8 is not supported.14

There is, however, an interaction between the two participation variables that parallels the interaction between gender of speaker and gender of follower. This interaction supports Hypothesis 9: The tendency for high-participation speakers to continue to develop their topics without experiencing topic loss is significantly stronger when the following speaker is also high participation. On the other hand, high-participation speakers are particularly likely to dismiss the topics of low-participation speakers. In other words, high-participation speakers discriminate more strongly between high- and low-expectation speakers in their topic transition behavior: They are more likely to make positive or negative judgments about other group members' contributions and to act on those judgments. This pattern is con- sistent with the finding that males are more likely to discriminate between male and fe- male speaker. In addition, this pattern ech- oes previous findings in which men are relatively unlikely to interrupt other men and are more likely to interrupt female speakers (Smith-Lovin and Brody 1986). Therefore, we are somewhat confident in regarding this pattern as a status phenom- enon. High-status or high-expectation people consistently discriminate more among their fellow group members than do low-status or low-expectation people. We should note, however, that the interaction coefficient was not significant in the dyad replication study. Evidently, this discrimi- nation phenomenon is more likely to occur

l4 Interestingly, the absence of a participation- of-follower effect parallels the findings in early conversation analytic studies, which found that the strongest effects are in topic loss versus de- velopment rather than in topic introduction. Fishman (1983), for example, found that men and women introduced similar numbers of topics, but that men's topics survived while women's failed.

in larger groups than in one-on-one mixed- sex conversations. l5

Looking at previous topic loss by the speaker and previous topic initiation by the follower, we see that these features of con- versation are repeated over the course of the conversation. That is, they form fairly stable conversational roles, and people tend to re- peat the same patterns that they have dis- played previously in the conversation. This pattern is consistent with our interpretation of topic transitions as power and prestige behaviors in task-oriented discussions, sup- porting Hypotheses 11 and 12. Suggesting new topics is a positive task cue that tends to mark a speaker as having a high-expecta- tion state; losing topics that one is develop- ing may be perceived as a negative task cue, indicating low status within the expectation structure of the group.16

TOPICCHANGE MARKERS. We examine two structural features of conversations that Maynard (1980) argues may mark that a topic is winding down or failing to create sufficient speaker interest to sustain normal turn-taking. We see that in both the six-per- son groups and in the dyads, a significant pause (.5 seconds or longer) increases the chance that the follower will introduce a new topic. In addition, an unusually short utter- ance just before a speaker transition in- creases the likelihood that the new speaker will introduce a different topic (although this effect is statistically significant only in the six-person groups).

Finally, there is a strong indication that speakers who originally introduced a topic have special rights to shift the topic to an- other matter of concern. We suspect that this pattern occurs because dismissing one's own topic is less likely to be taken as a negative evaluation, because the loss of one topic is balanced by the positive task cue of intro-

l5 We also checked for possible statistical in- teractions between the participation and gender variables. None of these interactions was signifi- cant.

l6 The topic history variables as well as the speaker's rights variable are not included in the analysis of the dyadic conversations because of the strong tendency to alternate topics in the dy- ads. In this case, the topic variables are so highly correlated that they cannot be analyzed sepa- rately.

ducing another (and both are indications of high participation and topic initiation).

TIME DEPENDENCE WITHIN TOPICS AND THE GROUP DISCUSSION. Topics tend to ei- ther be successful (last a long time) or to die out relatively quickly. The topic duration variable indicates that the longer a topic has been developed in the group discussion, the lower its hazard of dying. This effect is rep- licated in the dyad data; topics tend to be successful or to fall flat.

We also find that those topics initiated late in a group discussion are more likely to be successful (in gaining significant discussion and development) and are less likely to be lost. This time dependence across the entire conversation does not appear in the dyad data, possibly because the dyads are less constrained for floor time (as they were given the same 30-minute period to reach the collective task decision, but had fewer speakers from which to collect ideas).

The analyses in Table 3 combine topic tran- sitions by type: sudden, minimally linked, linked, and collaborative. Conversation ana- lysts have argued that men and women use these types of topic shifts with different fre- quencies and for different purposes. Expec- tation states theory also predicts some dif- ferences among the types of topic transitions (Hypotheses 4 and 8).

To test Hypotheses 4 and 8, we analyze the four types of topic transitions separately in a competing-risk framework for the six-person groups." In these models, linked and mini- mally linked transitions show substantively similar patterns of coefficients and do not differ significantly from one another. There- fore, we combine these two transition types into a single category of linked transitions in Table 4 (column 2).18

The results in Table 4 show some effects of gender, but they do not fall clearly into

l7 The number of transitions in the dyad data set is too small to allow stable analyses of transi- tion subtypes.

l8 This combination has the added benefit of increasing the number of cases for these smaller categories of transitions, making the power of this analysis similar to that of the collaborative and sudden transitions.

Table 4. The Effects of Gender, Participation, and Conversation Analysis Variables on the Three Types of Topic Transitions: Six-Person Groups
(1)     (2)     (3)
Independent     Collabor-        
Variable     ative     Linked     Sudden

Gender Variables

Gender of speaker

Gender of follower

Gender of speaker x Gender of follower

Group composition (no. of females)

Participation Variables

Participationof -.000 ,019 ,010 speaker (.OOO) (.012) (.010)

Participation of ,010 ,015 ,016' follower (.005) (.012) (.010)

Participation of -.001*** -.001*** -.000 speaker x (.OOO) (.OOO) (.OOO) Participation of follower

Previous topic loss .407*** .290*** -.023 by speaker (.062) (.066) (.035)

Previous topic ,443"' ,172"' -.046 initiation by (.043) (.048) (.035) follower

Topic Change Marker Variables

Pause -.I18 -.509 ,375 (.252) (.459) (.249)

Number of words -.033*** -.008 -.001 in previous turn (.004) (.005) (.005)

Speaker rights 2.16*** 2.80"' -.033 (.091) (.193) (.121)

Control Variables

Topic duration -.000 ,001*** -.001** since last transition (.001) (.001) (.001)

Time since -.346*** -.370*** -.I29 beginning of (.090) (.090) (.099) conversation (log)

Number of cases 235 236 377 Degrees of freedom 14 14 14 -2 log-likelihood 2,627.5 2,682.9 3,735.2 Chi-square 3,676.19 2,793.5 1 19.39

Note: Numbers in parentheses are estimated stan- dard errors. See Table 1 for description of data source.

*pc .05 **pc .O1 ***p< ,001 (two-tailed tests)

the pattern predicted by Hypothesis 4strong effects of gender of speaker, gender of follower, and their interaction for sudden transitions, intermediate effects for linked transitions, and least strong or even reversed effects for collaborative transitions. The re- sults providing strongest support for the hy- pothesis are the gender-of-speaker effects: Males are at less risk of losing their topics through a sudden topic transition (which is likely to be viewed as a negative comment on the topical content). On the other hand, men are more likely than women to experi- ence collaborative transitions. This might in- dicate that the status pattern is clearer for sudden transitions and becomes less strong (and even reverses) for collaborative transi- tions. The gender-of-follower and gender in- teraction variables do not show the same pat- tern, however. The gender-of-follower coef- ficient becomes stronger as we move from sudden to linked to collaborative topic changes: Men introduce topic changes through collaborative transitions at a consid- erably higher rate than do women, and this gender effect is smaller in linked transitions and nonsignificant in sudden transitions. The interaction term shows the same pattern, op- erating more strongly for the collaborative and linked transitions than for the sudden transitions.

The participation variables also fail to support Hypothesis 8, which predicted that low-participation speakers would be more likely to lose their topics than high-partici- pation speakers. The negative participation- of-speaker effect becomes nonsignificant and varies in sign when the analysis is split into transition types. The participation inter- action is stronger for linked and collabora- tive transitions than for sudden transitions (the opposite of what was predicted). Only the participation-of-follower variable shows a stronger effect for sudden transitions: It is significantly positive, indicating that high- participation speakers are more likely to in- troduce sudden topic shifts.

1 The number of females in the group has a negative effect on the risk of a sudden topic I transition (column 3), as Hypothesis 6 predicts. But the effect does not switch to a posi- tive impact in the collaborative topic transi- tions, as Hypothesis 5 predicts. Therefore, our conclusions about the impact of the num-

ber of women in the group on the group's culture or norms are tentative at best.

Some of the topic change marker variables show interesting patterns across the differ- ent types of topic transitions. Speakers who introduce a topic are no more likely than other group members to change that topic through a sudden transition; however, they are more likely than other group members to use linked or collaborative transitions to shift to a new subject or task solution. A minimal, short utterance appears to increase the risk of only collaborative of topic transi- tions, while pauses are less consistent mark- ers overall for transitions.

The coefficients for topic duration since last transition and time since the beginning of the conversation (log) are consistently negative and significant, showing that some topics are more successful than others, and that once a topic attains a certain level of the group's attention, the risk of a topic shift de- creases substantially. Similarly, task groups show a consistent pattern of trying out top- ics quickly at the beginning and settling on topics later in the group discussion for sus- tained development.



We began with the assumption that topic transitions would be indicators of the power and prestige order formed in task discus- sions. The significant effects of participation history and early topic-change behavior re- inforce our belief that topic transitions are an important part of the status structure in task-oriented conversations. Determining who directs the flow of conversation-indi- cating procedures that the group should fol- low, moving the group from one part of the process to the next, and suggesting potential solutions to the task-is structured in ways similar to other power and prestige behav- iors.

Initially, our primary focus on gender as a status characteristic, primarily because gen- der inequalities had been featured in much of the early conversation analytic work. How- ever, it now appears that topic shifts may be structured less by a relatively weak status characteristic like gender and may be more affected by endogenous processes through which people develop expectations for each other's competence within the context of the group. The effects of participation and ear- lier topic-change behavior seem to be more consistent and powerful factors than gender in these groups and dyads. Weak status char- acteristics may predispose certain conversa- tional patterns-receiving action opportuni- ties from others, taking the floor, receiving positive evaluations, and so on. However, prior successes within the group discussions and the content of topic suggestions and ref- erences to the legitimate structure of the task (as framed by the experimenters) may allow initially status-disadvantaged members to launch successful topics (especially if their status disadvantage is not too great). Because topic initiation and development may be col- laborative, and therefore easier to engage in legitimately, it may be a useful mechanism through which group members can contrib- ute to task solutions, even if they are not ini- tially high status.

Interestingly, our conclusions mirror re- cent discussions in both the conversation analytic tradition and in group processes theory. In the last decade, conversation ana- lysts have moved away from the study of how an isolated attribute like gender affects the frequencies of conversational behaviors (Gibson 1998; Schegloff 1987; Thorne et al. 1983). Instead, they have redirected their ef- forts toward the mechanisms through which conversations unfold in actual social con- texts. Similarly, the group processes researchers (in particular Skvoretz and Fararo 1996) have developed models of the proba- bilistic, sequential processes through which status structures affect behavior. These en- dogenous processes can reproduce exter- nally given status structures, but they also can easily overwhelm a weak status charac- teristic like gender in a specific group dis- cussion. We believe that our results show the importance of this new work. While not pro- viding the fine-grained description of con- versational mechanisms that a conversa- tional analysis would, we illustrate the im- portance of internal patterns within a conver- sation for shaping the relative influence of group members later in the conversation.

These endogenous processes may operate more strongly in our constrained task-ori- ented groups than in the less structured two- person conversations that conversation ana- lytic work has focused on. Naturally occur- ring conversations often find men and women in structural positions with status and power differences favoring the men. Kollock, Blumstein, and Schwartz (1985) found that power was more important than gender in determining the use of interrup- tions in conversations between domestic partners. By using experimental groups of previously unacquainted undergraduates, we minimized the important structural variation in men's and women's roles. In a sense we have examined gender as a pure status char- acteristic, holding constant to a large extent the structural forces that might have shaped the topic flow in the conversation analytic literature. In our experimental context, we find that gender has a subtle effect: Men ap- pear to discriminate and change the subject when it is being developed by women rather than other men. We see the same pattern in participation, with high-participation speak- ers often changing the subject developed by low-participation rather than high-participa- tion others. We conclude that gender and volubility may have subtle legitimacy ef- fects, enabling higher-status group members to evaluate the contributions of others and

act on those evaluations. These effects seem to operate more strongly in group contexts than in the dyads we studied.

Finally, we find that many observations about conversation sequences governing topic transitions that have been generated by the conversation analytic tradition also hold for the more constrained, task-oriented dis- cussions that we study here. Pauses and short, minimal turns at talk often precede topic changes. Sequences early in the con- versation-frequent topic initiation and par- ticipation-often prime later conversational behaviors. Although our event history analy- ses cannot approach the fine-grained analy- sis of conversational structures that charac- terizes the conversation analytic literature, they demonstrate the continuing synergy be- tween that tradition and the study of group processes. We expect this cross-fertilization to become even more important as the new work on the dynamics of group discussions and their endogenous status-organizing structures progresses (Fisek et al. 1991; Gibson 1998; Skvoretz and Fararo 1996).

Dina G. Okamoto is Assistant Professor of Soci- ology at the University of California, Davis. Her research interests include social psychology, so- cial inequality, and collective action. She con- tinues her research on panethnic group forma- tion among Asian Americans and on identity pro- cesses among racial and ethnic minorities in the United States.

Lynn Smith-Lovin is Professor of Sociology at the University of Arizona. Her research inter- ests include the relationships among identity, emotion, and action. She is beginning a study funded by the National Science Foundation on how people manage the identities of others with whom they interact.

Ainsworth-Vaughn, Nancy. 1992. "Topic Tran- sitions in Physician-Patient Interviews: Power, Gender, and Discourse Change." Language in Society 2 1 :409-26.

Allison, Paul D. 1984. "Event History Analysis: Regression for Longitudinal Event Data. " Sage University Paper Series on Quantitative Applications in the Social Sciences, no. 46. Newbury Park, CA: Sage.

Aries, Elizabeth. 1996. Men and Women in In- teraction: Reconsidering the Differences. Oxford, England: Oxford University Press.

Balkwell, James W. and Joseph Berger. 1996. "Gender, Status, and Behavior in Task Situa- tions." Social Psychology Quarterly 59:273


Bales, Robert F. 1951. Interaction Process Analysis: A Method for the Study of Small Groups. Cambridge, MA: Addison-Wesley.

Berger, Joseph, Bernard Cohen, and Morris Zelditch Jr. 1972. "Status Characteristics and Social Interaction." American Sociological Re- view 37:241-55.

Berger, Joseph, Thomas L. Conner, and M. Hamit Fisek. 1974. Expectation States Theory: A Theoretical Research Program. Cambridge, MA: Winthrop.

Berger, Joseph, M. Hamit Fisek, Robert Z. Norman, and Morris Zelditch Jr. 1977. Status Characteristics in Social Interaction: An Ex- pectation States Approach. New York: Elsevier.

Carli, Linda L. 1989. "Gender Differences in In- teraction Style and Influence." Journal of Per- sonality and Social Psychology 56:565-76.

Dovidio, John F., Clifford E. Brown, Karen Heltman, Steven L. Ellyson, and Caroline F. Keating. 1988. "Power Displays between Men and Women in Discussions of Gender-Linked Tasks: A Multichannel Study." Journal of Per- sonality and Social Psychology 55580-87.

Ellyson, Steven L., John F. Dovidio, and Clifford

E. Brown. 1992. "The Look of Power: Gender Differences and Similarities in Visual Domi- nance Behavior." Pp. 50-80 in Gender, Inter- action, and Inequality, edited by C. Ridgeway. New York: Springer-Verlag.

Fisek, M. Hamit. 1974. "A Model for the Evolu- tion of Status Structures in Task-Oriented Dis- cussion Groups." Pp. 53-84 in Expectation States Theory, edited by J. Berger, T.L. Conner, and M. H. Fisek. Washington, DC: Winthrop.

Fisek, M. Hamit, Joseph Berger, and Robert Z. Norman. 199 1. "Participation in Heteroge- neous and Homogeneous Groups: A Theoreti- cal Integration." American Journal of Sociol- ogy 97: 114-42.

Fishman, Pamela M. 1977. "Interactional Shitwork." Heresies: A Feminist Publication on Art and Politics 2:99-101.

. 1983. "Interaction: The Work Women Do." Pp 89-101 in Language, Gender, and So- ciety, edited by B. Thorne, C. Kramarae, and

N. Henley. Rowley, MA: Newbury House.

Gibson, David R. 1998. "Where Does Volubility Come From?" Paper presented at International Conference on Social Networks, May 25, Sitges, Spain.

Goffman, Erving. 1983. "Felicity's Condition." American Journal of Sociology 89: 1-53.

Goodwin, Marjorie Harness. 1980. "Directive- Response Speech Sequences in Girls' and Boys' Task Activities." Pp. 157-73 in Women and Language in Literature and Society, edited by S. McConnell-Ginet, R. Borker, and N. Furman. New York: Praeger.

James, Deborah and Sandra Clarke. 1988. "Women, Men, and Interruptions: A Critical Review." Pp. 231-80 in Gender and Conver- sational Interaction, edited by D. Tannen. New York: Oxford University Press.

Johnson, Cathryn. 1994. "Gender, Legitimate Authority, and Leader-Subordinate Conversa- tions." American Sociological Review 59: 122-


Johnson, Cathryn, Jody Clay-Warner, and Stephanie J. Funk. 1996. "Effects of Authority Structures and Gender on Interaction in Same-Sex Task Groups." Social Psychology Quarterly 59:22 1-36.

Kollock, Peter, Philip Blumstein, and Pepper Schwartz. 1985. "Sex and Power in Interac- tion: Conversational Privileges and Duties." American Sociological Review 50:34-46.

Lakoff, Robin. 1973. "Language and Woman's Place." Language and Society 2:45-80.

Maltz, Daniel N. and Ruth Borker. 1982. "A Cul- tural Approach to Male-Female Miscommuni- cation." Pp. 197-216 in Language and Social Identity, edited by J. J. Gumperz. Cambridge,

MA: Cambridge University Press. Maynard, Douglas. 1980. "Placement of Topic Changes in Conversation." Semiotica 30:263


Meeker, Barbara F. and P. A. Weitzel-O'Neill. 1977. "Sex Roles and Interpersonal Behavior in Task-Oriented Groups." American Socio- logical Review 42:91-105.

Ridgeway, Cecilia L. 1982. "Status in Groups: The Importance of Motivation." American So- ciological Review 47:76-88.

Ridgeway, Cecilia L. and Joseph Berger. 1986. "Expectations, Legitimation, and Dominance Behavior in Task Groups." American Socio- logical Review 5 1:603-17.

Ridgeway, Cecilia L., Joseph Berger, and LeRoy Smith. 1985. "Nonverbal Cues and Status: An Expectation States Approach." American Jour- nal of Sociology 90:955-78.

Ridgeway, Cecilia L. and Cathryn Johnson. 1990. "What Is the Relationship between Socioemotional Behavior and Status in Task Groups?" American Journal of Sociology 95: 1189-1212.

Ridgeway, Cecilia L. and Lynn Smith-Lovin. 1999. "Interaction in the Gender System: Theory and Research." Annual Review of So- ciology 25: 19 1-2 16

Ridgeway, Cecilia L. and Henry A. Walker. 1995. "Status Structures." Pp. 281-310 in Social Psychology: Sociological Perspectives, edited by K. S. Cook, G. A. Fine, and J. A. House. Boston, MA: Allyn and Bacon.

Riordan, Cornelius. 1983. "Sex as a General Sta- tus Characteristic." Social Psychology Quar- terly 46:261-67.

Robinson, Dawn T. and Lynn Smith-Lovin. 1990. "The Timing of Interruptions in Group Discus- sions." Advances in Group Processes 7:45-73.

Sacks, Harvey. 1967. Unpublished lecture, March 9. Transcribed by Gail Jefferson. School of Social Sciences, University of Cali- fornia, Irvine, CA.

Sacks, Harvey, Emanuel A. Schegloff, and Gail Jefferson. 1974. "A Simplest Systematics for the Organization of Turn-Taking for Conver- sation." Language 50:696-735.

Schegloff, Emanuel A. 1987. "Between Micro and Macro: Contexts and Other Connections." Pp. 207-34 in The Micro-Macro Link, edited by J. C. Alexander. Berkeley, CA: University of California Press. Schegloff, Emanuel A., Gail Jefferson, and Harvey Sacks. 1977. "The Preference for Self- Correction in the Organization of Repair in Conversation." Language 53:361-82. Skvoretz, John V. 1981. "Extending Expectation States Theory: Comparative Status Models of Participation in N-Person Groups." Social Forces 59:752-70.

Skvoretz, John and Thomas J. Fararo. 1996. "Sta- tus and Participation on Task Groups: A Dy- namic Network Model." American Journal of Sociology 101:1366-49.

Smith-Lovin, Lynn. 1990. "Transcripts of Six- Person Task-Oriented Groups: Small Group Discussions." Technical Report 90-9, Depart- ment of Sociology, Cornell University, Ithaca, NY.

Smith-Lovin, Lynn and Charles Brody. 1989. "Interruptions in Group Discussion: The Ef- fects of Gender and Group Composition." American Sociological Review 54:424-35.

Smith-Lovin, Lynn and Dawn Robinson. 1992. "Gender and Conversational Dynamics." Pp. 122-57 in Gender, Interaction, and Inequality, edited by C. L. Ridgeway. New York: Springer-Verlag. Smith-Lovin, Lynn, John V. Skvoretz, and Char- lotte G. Hudson. 1986. "Status and Participa- tion in Six-Person Groups: A Test of Skvoretz's Comparative Status Model." Social Forces 64:992-1005.

Tannen, Deborah. 1986. That's Not What I Meant! How Conversational Style Makes or Breaks Your Relations with Others. New York: William Morrow. . 1990a. "Gender Differences in Topical Coherence: Creating Involvement in Best Friends' Talk." Discourse Processes 13:73-90. . 1990b. You Just Don't Understand: Women and Men in Conversation. New York: William Morrow. Thorne, Barrie, Cheris Kramarae, and Nancy Henley. 1983. "Language, Gender and Society: Opening a Second Decade of Research." Pp.

7-24 in Language,Gender,and Society, edited by B. Thorne, C. Kramarae, and N. Henley. Rowley, MA: Newbury House.

Wagner, David G. and Joseph Berger. 1997. "Gender and Interpersonal Task Behaviors: Status Expectation Accounts." Sociological Perspectives 40: 1-32.

West, Candace. 1979. "Against Our Will: Male Interruptions of Females in Cross-Sex Conver- sation." Language, Sex, and Gender: Annals of the New York Academy of Sciences 327:81-97.

West, Candace and Angela Garcia. 1988. "Con- versational Shift Work: A Study of Topical Transitions between Women and Men." Social Problems 35:550-75.

West, Candace and Don H. Zimmerman. 1983. "Small Insults: A Study of Interruptions in Cross-Sex Conversations between Unacquainted Persons." Pp. 103-17 in Language, Gender, and Society, edited by B. Thorne, C. Kramarae, and N. Henley. Rowley, MA: Newbury House.

Wood, Wendy and Stephen J. Karten. 1986. "Sex Differences in Interaction Style as a Product of Perceived Sex Differences in Competence." Journal of Personality and Social Psychology 50:341-47. Yamaguchi, Kazuo. 1991. Event History Analy- sis. Applied Social Research Method Series, vol. 28. Newbury Park, CA: Sage. Zimmerman, Don H. and Candace West. 1975. "Sex Roles, Interruptions and Silences in Con- versation." Pp. 105-29 in Language and Sex: Difference and Dominance, edited by B. Thorne and N. Henley. Rowley, MA: Newbury House.

  • Recommend Us