INTERRUPTIONS IN GROUP DISCUSSIONS: THE EFFECTS OF GENDER AND GROUP COMPOSITION*

LYNNSMITH-LOVIN CHARLESBRODY Cornell University Tulane University Conversations both reflect and maintain social inequalities. They import hierarchical structures from larger society and help perpetuate them by creating inequalities in the ability to accomplish interactional goals. In this study of speaker transitions in six-person, task-oriented experimental groups, we explore the well-known finding that men interrupt women more frequently than women interrupt men. We ask three questions about the structure of interruptions. Who attempts to interrupt whom and under what conditions? How does the affective character of interruptions vary across speakers and groups? What determines whether an interruption succeeds? We find that gender inequality in these task-oriented discussions is created by a mixture of attempts to use power and of differential success. In their interruptions, men discriminate by sex in attempts and in yielding to interruptions by others. Women interrupt and yield the jloor to males and females equally. The sex composition of the group affects interruption patterns in complex ways. Men interrupt men with supportive comments in all-male groups, but these supportive interruptions drop as the number of women in the group increases. Supportive interruptions also succeed in gaining the jloor more often in single-sex groups. Taken together, the results suggest a mixture of status and conjlict models and reaflrm the importance of group composition in interaction.

Studies of group discussions reveal that High-status actors talk more, are more status1 has many effects on participation. successful at introducing topics, interrupt High-status people are asked their opinions more, and receive more positive feedback more often, talk more, receive more positive from their listeners. comments, are chosen as leader more fre- Conversation is an important domain for quently, are more likely to 'influence the studying these status effects. Language use groups' decisions, and in general dominate the conversation (see review in Ridgeway can tell us a great deal about the nature and 1983, pp. 160-204). Status differentiation not extent of social inequality. It reflects the only creates conversational dominance, but hierarchical social structure outside the group, often legitimates it (Ridgeway and Berger while simultaneously providing the means 1986). Studies of conversations in dyads or through which that inequality is maintained. families also frequently find status effects. Differences by status in participation, conver- sational dominance, and adherence to turn- taking norms create differences in our ability * The authors thank Paul Benson, Ron Breiger, to get our ideas across to others. In a family Fred Koenig, Bruce Mayhew, James McRae, Miller McPherson, Victor Nee, Patrick Nolan, this may mean control over decision making; Mark Pltmkett, Cecilia Ridgeway, Joseph Sheley, in a work setting it may determine perfor- John Skvoretz, David Weakliem, and two anony- mance and promotion. mous reviewers for helpful comments on the In this study, we are interested in the status paper. They also thank Charlotte G. Hudson for allowing them to use the data, which were differences between males and females. There collected for another purpose in her master's thesis is extensive evidence that sex operates as a research. Brody's work was supported partially by status characteristic in small groups research NIMH grant R01-MH40666. The order of the (e.g., Strodbeck, James, and Hawkins 1957), authors' names was determined by a coin toss. in the expectations states paradigm (e.g., ' Following Ridgeway (1983, p. 160), we define a member's status as "the degree of Wagner, Ford, and Ford 1986; Lockheed deference, esteem and power to influence others" 1985), and in studies of conversation (e.g., he or she acquires. Kollock, Blumstein, and Schwartz 1985; 424 American Sociological Review, 1989, Vol. 54 (June:424-435) INTERRUPTIONS IN GROUP DISCUSSION West and Zimmerman 1977).2 In group high-status speakers more access to important discussions, men talk more, more often interpersonal resources (the "floor") at the assume a leadership position, receive more expense of their lower-status partners. They positive statements and fewer negative state- may also serve to disorganize the speech and ments, are more likely to show nonverbal task ideas of the interrupted (West 1979). As and dominance cues, and so on. Kollock et al. (1985, p. 40) argued, attempt- The aspect of discussion in which we are ing to interrupt is an excellent indicator of interested is interrupti~n.~An interruption attempted conversational control; successful occurs when one speaker disrupts the turn of interruptions are a sensitive measure of actual another with a new utterance. Normally, dominance. Power is a human accomplish- conversation is organized so that one speaker ment, situated in everyday interaction; inter- talks at a time; speakers alternate in turns to ruptions are one of the mechanisms that prevent the conservation from becoming a accomplish power in discussions. monologue (Sacks, Schegloff, and Jefferson Unfortunately, there are numerous prob- 1974). Speakers indicate when their turn is lems in studying interruptions. First, they are coming to an end with a variety of nonverbal relatively rare events. Most conversation is and tonal cues. Typically, turn taking in very well coordinated. In their classic study conversations is achieved without substantial of naturally occurring conversations, Zimmer- gaps or overlaps in speech (Dabbs, Ruback, man and West (1975) found only 7 interrup- and Evans 1985). As Kollock et al. (1985) tions in 20 same-sex two-party conversations note, a speaker's turn is his or her opportunity and 48 interruptions in 11 cross-sex conversa- to accomplish interactional goals or to block tions. Drass (1986) found 61 interruptions in others from accomplishing theirs. To interrupt 28 single-sex conversations-only slightly a speaker-that is, to begin talking before the more than 2 interruptions per conversation. In speaker's turn is finished and perhaps prevent a more recent study of couples, Kollock et al. him or her from completing the thought-is to (1985) found an average of 18.6 attempted prevent, at least temporarily, that speaker interruptions per quarter hour of conversation, from accomplishing these interactional goals. less than half of which were successful. Since Since interruptions represent a clear viola- interruptions are infrequent, we need long tion of turn-taking norms that give one specimens of conversation or large samples of conversant greater access to others' attention, interactions to get stable estimates. The we are not surprised that their occurrence is skewed distribution of transitions into normal linked to dominance, power, and status (see, versus interrupted categories also causes e.g., Drass 1986; Eakins and Eakins 1978; problems for statistical analyses. Zimmerman and West 1975; West 1984). A second problem arises from the fact that Earlier studies have found that men interrupt one must speak before one can be interrupted. women, adults interrupt children, doctors Even in a two-person conversation, the two interrupt patients (except when the doctor is a interactants are not equally at risk of being "lady"), the more powerful spouse interrupts interrupted. The person who talks more is at the less powerful one, and those with risk more of the time. When more than two masculine identities interrupt those with more people participate in a discussion, the prob- feminine self-images. These violations of lem will be exaggerated since participation is turn-taking norms clearly allow the powerful, known to be strongly skewed (Horvath 1965). The risk factor is especially important since we know that participation is status and power Some studies raise questions about the univer- dependent (Strodbeck et al. 1957; Berger, sality of gender as a status characteristic in our Rosenholtz, and Zelditch 1980). An addi- culture. However, earlier analyses (Smith-Lovin et tional risk factor involves the success of al. 1986) of University of South Carolina student interruption attempts. An interruption that data indicated that sex is a status characteristic in succeeds in breaking off a speaker's turn may this population. be more disruptive than one that the speaker We focus on only half of the conversational process, ignoring the important "listener" role and manages to fend off. But an interruption must the interactional work done by the nonspeaker. In be attempted before it can succeed. There- our six-person discussion groups, speaking acts fore, the issue of interruption attempts is could be coded much more reliably than the subtler somewhat different from that of interruption listener responses. success. AMERICAN SOCIOLOGICAL REVIEW A third problem is that interruptions may intimate dyads is likely to underestimate differ in their impact depending on their interruption rates and their sex differentiation. content. While an interruption that comments In studies of larger, less intimate groups, positively on an idea, repeats a phrase, or group composition variables become impor- finishes a thought for a speaker is clearly a tant. ~0th-researchersand popular writers violation of turn-taking norms, it is support- have noted that groups of different sex ive of the speaker and may be the basis for composition have different patterns of talk. highly collaborative talk (Tannen 1987). All male groups seem to be more competitive, Interruptions that interject negative comments to form dominance hierarchies more quickly, or put-downs or completely ignore the and to maintain stable hierarchies during speaker by introducing an unrelated topic are continued interaction (Aries 1976). Some more intrusive and disruptive; they are much evidence suggests that both men and women more explicitly an attempt at power. More express more negative and less positive neutral interruptions may be related to issues socioemotional behavior in mixed-sex than in of power or may result from lack of a shared same-sex groups (Anderson and Blanchard speech style. Lumping types of interruptions 1982). into a single analysis may mask important The best developed theoretical work on the differences in conversational dynamics. group composition question is Kanter's (1977) A final concern is the importance of social theory of proportional representation. Kanter context in determining conversational interac- suggests that variations in sex composition tion. As Thorne, Kramarae, and Henley within the mixed-sex category are important. (1983, p. 13) noted, the study of isolated In particular, minority group members are conversational variables almost universally under special interaction pressures in groups leads to further questions about the effects of with a skewed distribution. Such token setting, topic, roles, and other social factors members may adopt conservative behavior that may interact with gender. Three factors patterns, avoid conflict, and become a that might affect interruptions are considered relatively passive audience for the boundary- here: the intimacy of the conversants, the size heightening behavior of the majority. Empir- of the group, and the sex composition of the ical studies generally have supported Kanter's group. ideas, showing that group dynamics and minority achievement differ in skewed as opposed to more representative groups (e. g , THE CONTEXT OF CONVERSATION . Spangler, Gordon, and Pipkin 1978; Alex- The research on interruptions has concen- ander and Thoits 1985). South, Markham, trated heavily on conversations in intimate Bonjean, and Corder (1987) and Wharton and dyads. While these intimate settings are Baron (1987) suggest that social support and important domains for the exercise of interper- interaction quality differ in work groups with sonal status and power, these studies may not varying sex composition. generalize to a wide variety of more imper- We have two goals for the current analysis sonal, task-oriented situations. Several re- regarding the context effects on interactional searchers have found that intimacy influences process. First, we widen the context of the conversational dynamics. Intimates have less earlier research on interruptions by studying sex-differentiated conversations than noninti- interactions among nonintimates in a task- mates (Heiss 1962). Among friends, overlaps oriented setting. Second, by using larger in speech are frequently instances of highly groups and systematically varying sex compo- collaborative talk: cooperative sentence build- sition, we study the effects of this complex ing, requesting and giving verification through context variable on the interaction. backchannels, and "savoring repetitions" (Tannen 1983). But friends also interrupt less INTERRUPTIONS IN GROUP and have more silences (Shaw and Sadler DISCUSSIONS 1965). Among stable couples, power and dominance are likely to be more important In this study, the task-oriented nature of the than sex in determining interruptions, ques- groups serves two purposes. It creates an tion asking, and topic control (Kollock et al. instrumental framework similar to that of 1985; Vuchinich 1984, p. 23 1). Therefore, many work groups, allowing us to study a the typical study of interruptions among domain iri which the ability to promote ideas INTERRUPTIONS IN GROUP DISCUSSIONS 427 and gain recognition is likely to be particu- different types of interruptions work for men larly important. It also produces conditions and women (or against men and women)? under which status effects are most likely to Such questions are important, for these be found (Berger et al. 1980). successful interruptions are the most serious We use six-person groups both to create the intrusions into speakers' talk. They represent differentiation that is typical of larger groups conversational dominance -the successful use and to have groups in which Kanter's of power in these group discussions. definition of a skewed group could be met To control for the frequency with which (with a 1:5 ratio). The sex composition of the males and females are at risk of being groups systematically varies, while variation interrupted and to allow more detailed in other status characteristics is minirni~ed.~statements about the pattern of interruption Group members' prior knowledge of each attempts and their success, we analyze the log other was minimal5; our assumption was that odds that an interruption was attempted this lack of previous intimacy will heighten (relative to normal, uninterrupted transitions), the effects of the observable status character- as well as the log odds of an attempted istic (gender) relative to other status or power interruption succeeding.6 Such an analysis differences that might develop as the result of controls for risk factors; if males participated long-term interaction. more frequently than females (and were at We address three questions about interrup- greater risk of being interrupted), there would tions and conversational dynamics. First, who simply be more conversational transitions attempts to interrupt whom? Attempts to involving males. The relative size of this interrupt represent disregard for the speaker marginal would not affect the estimates of the and an attempt to block his or her interac- degree to which sex and group composition tional goals. We analyze who makes these influence interruptions. This form of analysis attempts at conversational dominance, who also compensates for the fact that different are the recipients of these power plays, and group compositions offer differing opportu- how the group context of the discussion nity structures for maletfemale intera6tion affects power attempts. (e.g., in a group with two females and four While any interruption attempt violates males, the females would be more likely to turn-taking norms, some interruptions may be interrupt a male than a female simply by more intrusive than others. We ask how the change -a given female would have only one character of interruptions varies across speak- other female to interrupt as opposed to four ers and groups. For example, are men or males). women more likely to make supportive interruptions? Do certain group compositions create more negative interruptions? A logit model was used in preference to other Finally, what determines whether an inter- statistical techniques because interrupted transi- ruption succeeds? Are some interrupters more tions are rare relative to normal transitions likely to wrest the floor from another? Does (including overlaps, simultaneous speech, unsuc- this success rate vary depending on the cessful interruptions). Since the transitions are the affective character of the interruption? Do unit of analysis here, the statistical model requires the unlikely assumption that each transition is an independent event. Kollock et al. (1985, p. 39) Students who wanted to participate filled out note that the behavior of an individual within a an index card with their name, social security couple is not independent of his or her partner; the number, age, and major area of study. The sex and same is undoubtedly true for group members' race of students were noted surreptitiously as the behavior within a group-it will be influenced by cards were collected. The cards were used to other members and by previous features of the manipulate sex composition, to hold constant other conversations. Note, however, that Vuchinich status characteristics (age, major, race), and to (1984) finds in family discussions that sequencing assign students randomly to groups within the in oppositional moves can be described as a li,mits imposed by sex composition and class times. first-order Markov process: one move affects the Students from eight different classes (most next, but not a long chain of consequences. Since with more than 50 students enrolled) constituted the independence assumptions of the statistical the subject pool. Postexperimental questionnaires model are not perfectly met, we regard all confirmed that the randomized selection of stu- significance tests to be suggestive only. In effect, dents from large classes effectively eliminated we used such tests to help locate patterns in the prior personal knowledge in most cases. data rather than in a hypothesis-testing framework. AMERICAN SOCIOLOGICAL REVIEW To address the important questions about place despite the intrusion, we considered the the affective character of interruptions, we interruption attempt to be unsuccessful. For analyse the odds that an attempted interrup- purposes of the analysis here, a noninter- tion is supportive or negative (as contrasted rupted transition constituted all other occa- with neutral comments and elaborations). At sions when one speaker followed another a later stage of the analysis, we also look at (including minor overlaps of speech and the odds of success separately for supportive, normal transitions). Simultaneous starts were negative, and neutral interruptions. rare; although the continuation of speech after a simultaneous start may be an indicator of dominance, it is not a violation of turn-taking DATA COLLECTION7 norms. Therefore, we did not code such The subjects were undergraduates enrolled in simultaneous starts as an interrupted transition introductory Sociology. The experiment used for this analysis. Similarly, we did not code only white students between the ages of 17 backchannel utterances that produced simulta- and 25 to minimize variation in important neous speech as attempted interruptions; statuses other than gender. We systematically Tannen (1983) argues convincingly that such varied the sex composition of the six-person overlaps are not norm violations and that they groups, with approximately five groups repre- support rather than disrupt the speaker's senting each of seven experimental conditions utterance in most cases. (see Table I).* After six months, we recoded one group's A sex-neutral collectively oriented task (a videotapes to check coder reliability. Of 186 variation of the task developed by Fisek 1974) transitions, the coder credited 8 different stimulated an active group discussion. The speakers in the two codings, resulting in an task is described more completely in Srnith- accuracy of 96 percent. In none of these cases Lovin, Skvoretz, and Hudson (1986). To was an interaction originally credited to one motivate the groups, we awarded a $30.00 sex shifted to another; nor was the interrupted prize ($5.00 per group member) at the end of versus noninterrupted status of a transition the semester to the group who arrived at the changed. best solution. To minimize confounding We coded the affective character of inter- effects of seating, we placed the six group ruptions into three categories using typed members around a circular table with ran- transcripts of the conversations. We consid- domly assigned seating. ered an interruption supportive if it expressed Two video cameras recorded interaction agreement with the speaker, if it made an through one-way mirrors. We later tran- affectively positive request for elaboration scribed all group discussions, using the best ("that's a great idea; how do you think we videotapes as a primary source, while relying should do it"), or if it completed the on the second to resolve problematic ex- speaker's thought (often repeating several of changes. We coded an interruption attempt the speaker's words, then continuing to a when one speaker's utterance was broken by normal transition place). We coded an an intrusion into the internal structure of the interruption as negative if it expressed speech act, not corresponding to a possible disagreement with the speaker, raised an transition place (following the definition by objection to the speaker's idea, or introduced West and Zirnrnerman 1977). We coded an a complete change of topic (disregarding the interruption attempt as successful when one speaker's utterance entirely). We coded all speaker's utterance prevented another speaker other interruptions as neutral. This category from completing a speech turn. If the speaker included interruption attempts that were so completed the utterance to a normal transition short that their content could not be deter- mined, interruptions that elaborated on the topic of the interrupted speaker without A more complete description of data collection evaluative content, and requests for clarifica- procedures is available in Smith-Lovin et al. 1986. Data on 31 groups in all were collected. Five tion where it was unclear from context groups were studied in all seven conditions except whether the request was a challenge or a all female (four groups), five females (three request for more talk. We do not imply by our groups), and four females (four groups). Schedul- labels that supportive interruptions are not ing problems prevented these conditions from disruptive to a speaker's talk and his or her having five groups. ability to accomplish interactional goals. In INTERRUPTIONS IN GROUP DISCUSSIONS 429 fact, one reader has suggested that in followers that are logically impossible. Thus, established power structures such supportive six rows are completely blanked since, for interruptions can be used to signal power; the example, it is impossible for a male to follow high-status actors may use positive comments a female speaker in an all-male or all-female to reward task-related behaviors from subor- group. Rows 8 and 21 are only partially dinates or to take over the discussion of blanked. Male-male transitions are possible in promising ideas suggested by others. How- groups with only one male member, as are ever, since such supportive disruptions may female-female transitions in groups with only spring from different conditions and lead to one female member. Such transitions were differing consequences than the more conflict- coded whenever a pause of greater than 1.5 ual negative interruptions, we distinguish seconds occurred between two utterances them for further analysis. made by the token member of the group (following Fisek's [I9741 definition of an action opportunity). Such transitions, how- ANALYSIS AND RESULTS ever, cannot be interruptions. We analyze speech transitions in the form of Several observations about group interac- an incomplete 7 X 2 X 2 X 7 contingency tion are immediately apparent. First, intenup- table (see Table 1). This table includes eight tions are very rare: 540 interruptions com- rows with structural zeroes (indicated by *) pared to 4,576 noninterrupted transitions in which represent combinations of speakers and the 3 1 group discussions. Group discussions

Table 1. Outcome of Speech Act by Sex of Speaker, Sex of Follower, and Number of Females in Six-Person Groups Outcome of Speech Act Interrupted Successful Not Successful Number of Sex of Sex of Not Females Speaker Follower Supportive Negative Neutral Supportive Negative Neutral Interrupted 0 M M 16 0 5 7 3 4 534 0 M F * * * * * * * 0 F M * * * * * * * 0 F F * * * * * * * 1 M M 9 3 14 15 3 15 674 1 M F 0 0 1 3 3 3 67 1 F M 2 4 6 1 1 3 61 1 F F * * * * * * 14 2 M M 4 2 9 5 5 8 380 2 M F 1 2 6 3 0 12 192 2 F M 1 5 6 6 1 10 187 2 F F 4 0 10 4 0 4 116 3 M M 4 2 3 4 1 0 198 3 M F 4 1 5 4 2 16 160 3 F M 5 3 9 1 3 2 145 3 F F 3 0 1 3 1 0 103 4 M M 0 1 4 0 0 3 118 4 M F 7 1 7 4 3 10 220 4 F M 3 2 12 4 2 11 213 4 F F 3 4 11 3 1 9 24 1 5 M M * * * * * * 24 5 M F 3 1 2 0 2 5 120 5 F M 1 1 4 1 2 5 111 5 F F 7 5 8 4 2 14 265 6 M M * * * * * * * 6 M F * * * * * * * 6 F M * * * * * * * 6 F F 13 12 15 5 3 14 43 3 - - * Denotes a structural zero (a combination that is logically impossible). Note that groups with only one female can have a female speech act followed by another female speech act and groups with only one male can fave a male-male transition. Such occurrences are the result of a pause greater than 1.5 seconds between two utterances made by the "token" member of the group. AMERICAN SOCIOLOGICAL REVIEW typically lasted 15 minutes, with approxi- variable. We also considered similar series of mately 17.4 interruption attempts per quarter models in which the effect of number of hour. This rate is similar to the 18.6 attempts females in the group was constrained to per quarter hour observed by Kollock et al. linearity. (1985). Explicitly negative interruptions are especially rare in our data, marring only 1.79 percent of the observed speaker turns. Interruption Attempts We employed a variant of the continuation- odds strategy suggested by Fienberg (1980) to To address the first question of who attempts analyze Table 1. We first considered the odds to interrupt whom and under what conditions, that an interruption attempt is made. We also we combine the first six columns of Table 1, analyzed the relative odds of supportive, contrasting all six types of attempted interrup- negative, and neutral attempts. In a later stage tions with uninterrupted transitions. Our of the analysis, we focused only on the preferred model includes an interactive effect attempted interruptions, analyzing the odds of of the sex of speaker and the sex of follower. success. Thus, the analyses of Table 1 The odds of a male attempting to interrupt proceeded by focusing on 'several partitions of another male (.078) are less than one-half the the seven columns, which correspond closely odds of a male attempting to interrupt a to the questions raised above. female (.163). Females attempt to interrupt At each stage of the analysis, we estimated male and female speakers at essentially a set of logit models. Since the odds that we identical levels (odds of .I46 and .I41 described above are asymptotically indepen- respectively). In other words, men discrirni- dent (Feinberg 1977, p. 87), we estimated the nate in their interruption attempts, interrupt- models and assessed their fit separately at ing women much more often than men. each stage of the analysis. The individual Women, on the other hand, do not discrimi- chi-squares were then added to get an overall nate; they attempt to interrupt men and goodness-of-fit measure for the set of models women equally. Group composition does not (see Table 2). appear to influence interruption attempts. Net The total baseline chi-square represents the of the sex of speakerlsex of follower model of independence of the row and interaction, neither the linear nor uncon- column classifications of Table 1. In Table 2, strained effects of number of females are this model yields L2 = 219.3 1, df = 114. The significant. In particular, we do not see any overall fit for our set of preferred9 models is impact of token status on interruption at- given by L2 = 126.78, df = 104. Since all our tempts in the (15) and (5:l) groups. Group models are logit-specification models which composition does not interact with other include the highest-order interaction among independent variables in affecting attempted the independent variables, the eight rows of interruption. the table including structural zeroes are effectively eliminated from the computations. Each partition of Table 1 differs only in the The Affective Character of numbers of columns (i.e., the aspect of the Interruption Attempts interruption variable that is analyzed). Hence, degrees of freedom for each baseline model How does the affective character of interrup- can be calculated as (20 - l)(#columns tion attempts vary across speakers, interrupt- - 1). We began the analysis of each partition ers, and groups? To address this question, we by estimating the standard logit models for a combine successful and unsuccessful interrup- four-way table, testing for various combina- tion attempts, retaining the distinction be- tions of main and interactive effects of sex of tween positive, negative, and neutral charac- speaker, sex of follower, and number of ter. Thus, the dependent variable has three females in the group on the interruption categories: supportive, negative, and neutral interruption attempts. The results of this We use the term "preferred model" below to analysis indicate a significant four-way inter- indicate a model that cannot be simplified without action among sex of speaker, sex of follower, a significant increase in the likelihood-ration number of females, and the character of the chi-square and cannot be significantly improved on interruption. Inspection of the observed odds by including additional effects. suggests a concise representation for this INTERRUPTIONS IN GROUP DISCUSSIONS 43 1 Table 2. Fit of Baseline and Preferred Model for Each Partition of Table 1

Baseline Model Preferred Model Partition df L* df LZ A. Interruption attempts: (attempted interruption vs. normal transition) 19 60.42 16 15.97 B. Content of interruption attempts: (supportive vs. negative vs. neutral) 38 74.42 36 51.87 C. Success of interruptions: i. positive 19 26.29 18 19.91 ii. negative 19 30.08 18 21.24 iii. neutral 19 28.10 16 17.79 TOTAL 114 219.31 104 126.78 interaction. This preferred modello includes a the linear relationship described above. Simi- linear effect of group sex composition (number larly, the groups containing tokens did not of females) only for male attempts to interrupt influence the odds of an interruption being other males with supportive (as opposed to supportive or neutral (as opposed to negative) neutral or negative) comments. This effect is in affective character. negative, indicating that males direct more supportive interruption attempts toward other males in all-male groups (a fitted odds of Success of Interruptions 1.82). The odds of a male-male interruption being supportive decline steeply as the What factors determine whether an interrup- proportion of women in the group increases, tion succeeds? To address this question, we reaching a low of .26 in groups with just two originally analyzed the odds of an interruption males. For all other sex combinations (male- attempt succeeding by affective character of female, female-male, and female-female), the interruption, by sex of speaker, by sex of odds that an interruption attempt is positive follower, and by group composition. That are equal (.47) and do not vary with group analysis showed that the answer to the composition. Under this model, the odds that question of who is more likely to succeed an interruption. attempt is negative (.30) also varies by the affective character of the are not affected by group composition or attempt. Thus, in this section we model the gender. odds of success separately for positive, Our analyses revealed no token effect in negative, and neutral interruption attempts. support of Kanter's ideas." The (15) and For positive interruption attempts, we find (5: 1) groups did not deviate significantly from a significant nonlinear effect of group compo- sition on the odds- of success. In mixed sex groups, positive interruption attempts have 'O Examination of residuals indicates that the about a 50-50 chance of success (an odds of relative lack of fit for this portion of our model is .94). Positive attempts are more than twice as attributable to two cells in the cross-classification- those pertaining to same-sex (male-male and likely to succeed in all-male or all-female female-female) supportive interruption attempts in groups (odds of 2.42). Men and women who groups containing three males and three females. are faced with a supportive intrusion in such Our preferred model underestimates the frequen- groups yield the floor much more frequently cies in these two cells; reestimating the model than their counterparts with an opposite-sex while "blanking" these cells results in L~=32.61, audience. Groups with a token composition df= 34. However, since the combined frequency in (15 and 5:l) did not differ significantly from these two cells is just 14 and since we have no other mixed-sex groups (with 2:4, 3:3, and compelling theoretical reason to expect such a 4:2 distributions of males and females). pattern, we prefer the model described here. For the relatively rare negative interruption 11 While our ideas have been directed by Kanter's discussion, this analysis cannot be attempts, the predicted odds of success considered a test of her theory of proportional depend only on the sex of the person being representation. The link between the outcomes that interrupted. The odds of a man yielding the she discusses and the conversational processes of floor to a negative interruption are .59; for transition and interruption is too loose to be more women, these odds are over three times as than exploratory. great (2.24). There is no significant effect of AMERICAN SOCIOLOGICAL REVIEW the sex of the interrupter-both men and more able to fend off potential disruptions, women are more likely to succeed in especially when directed at them by women interrupting the speech of women with (this pattern is especially clear with neutral negative comments. interruptions). Finally, for neutral interruptions (the most In effect, men are acting as if sex is a status frequent category) we find significant main characteristic (within the expectation states effects of both sex of speaker and sex of framework) (Berger et al. 1980) but women follower and a marginally significant (g = .075) are not. Men interrupt women more than interactive effect of these two variables on the other men, a pattern we would expect if they odds of success. Under the simpler model, had higher task expectations for males than women succeed less frequently than men females. when they attempt to interrupt with a neutral By attempting to interrupt men at the same comment, while they are more likely to lose rate as other women, and by yielding to the floor when someone attempts to interrupt intrusions at the same rate whether they come them. The interactive effect clarifies this from men or women, women are behaving as result. When men are speaking, women are though sex were not a status characteristic for less than half as likely as men to successfully them. While it is possible that the statuses of interrupt them with a neutral, elaborating male and female are linked to performance comment (.45 compared with 1.17). When expectations for males and not for females, it women are speaking, the odds of success are seems unlikely that the two sexes could roughly the same for women and men (1.19 maintain such divergent performance expecta- and 1.10). Therefore, the exceptional ability tions during long-term interactions. Perhaps of men to hold the floor against a female's this is why Kollock et al. (1985) found that neutral interruption is the major feature power in the relationship is more important creating the two main effects observed in the than gender among established couples. simpler model. Still, it is difficult to believe that men and The success of negative and neutral inter- women are bringing radically different ideas ruption attempts does not depend on group about male and female competence to our composition. Most importantly, we find no small-groups setting. Many expectation states evidence that token members of a group are experiments have shown that sex is a status more or less successful in their interruption characteristic for both men and women and in attempts; nor are they more or less likely to our society (see review in Wagner et al. yield the floor. 1986). Furthermore, in other analyses of these data we observe the participation inequalities by sex that typically indicate DISCUSSION status ordering (Smith-Lovin et al. 1986). It is This analysis of interruptions in large, unlikely that men and women have substan- task-oriented discussion groups supports pre- tially different views of the societal status vious research and adds new information structure. about the form of the relationships. We find Another view that could explain the pattern support for the typical finding that there is would be a mixture of status effects and of considerable gender inequality in interrup- conflict between males and females within the tions. But we see that this inequality is not the groups. In this combination, both status result of a sex difference in frequency of differences and adversarial relations between interruption, or even a simple dominance males and females would lead males to effect. Instead, it is produced because men interrupt women more than other men. But discriminate in their interruption attempts, for females, the two effects would produce disrupting the speech of women far more opposite predictions: status differences would frequently than that of men, while women do suggest deferring to men while interrupting not discriminate, interrupting women and other women, while adversarial conflict men equally often. This pattern carries over would suggest interrupting men while respect- into the pattern of successful interruptions, ing or supporting the speech of other women. since interruptions are more likely to succeed If the two effects were roughly equal in against women than against men (especially strength, they could cancel one another, when interruptions are disruptive and negative producing no sex difference in interruptions in character). Men, on the other hand, are for females. INTERRUPTIONS IN GROUP DISCUSSIOP Several recent studies yield some indirect ruptions are also more likely to succeed in a support for conflict dynamics in mixed-sex single-sex context. groups. Ridgeway's (198 1, 1982) experi- These indications that single-sex groups ments show that females in mixed-sex groups have a more supportive, lively, non-status- lose favor quickly if they do not express a dominated flow of ideas is supported by some group orientation (through group-oriented work in the Bales tradition (Anderson and cues and conformity). Taps and Martin Blanchard 1982). This research indicates both (1988) find that females who are tokens in men and women express somewhat less all-male groups need external accounts for positive socioemotional behavior in mixed- their ideas to gain influence; they also find sex than in single-sex groups. Kanter (1977) that women are less liked in sex-balanced also points to some maladaptive consequences groups than in single-sex or token groups. of interaction in mixed groups, especially in South et al.' (1987) show that both men and minority-majority interactions within skewed women get less social support for achieve- groups. (However, in our data, skewed ment from their opposite-sex coworkers when groups do not differ from more numerically the work group becomes evenly mixed. balanced compositions.) Similarly, Wharton and Baron (1987) find Perhaps positive, spontaneous, and pleas- that men have lower job satisfaction, lower antly disorganized speech occurs most fre- self-esteem, and higher job-related depression quently between status equals. When status when they are in evenly sex-mixed work differentiation and/or opposing group member- settings (as opposed to either male-dominated ships are activated in more heterogeneous or female-dominated settings). They interpret groups, much of the supportive interruption these findings to indicate a decline in the drops out (or succeeds less frequently). quality and quantity of intergroup relations in Perhaps the presence of a common speech such numerically balanced settings. style in a homogeneous group leads to more While all of the above studies could be relaxed, less conflictual group interaction. interpreted in different ways, they are all Tannen (1980) suggested that speakers may consistent with cross-sex conflict in groups. If be characterized by stylistic patterns (high- we add this conflict imagery to the status involvement versus high-considerateness) that differentiation between males and females include features such as frequency of overlap, that is so widely demonstrated in the early length of pause between utterances, prefer- small groups literature and in the expectations ence for shifts in amplitude and pitch, and states tradition, we may explain the sex tendency to talk through overlaps. When differentiation in males' interruption patterns these speech styles are shared, they provide and the apparent lack of discrimination in the basis for well-coordinated talk. However, females' interruption attempting and yielding when opposing styles are mixed in conversa- patterns (as the status and conflict effects tional groups, negative interpretations of cancel one another out). others' actions are frequent. (High-involve- The complex effects of group composition ment speakers think slower-paced talk indi- and the affective character of interruptions on cates lack of rapport and interest; high- conversational dynamics in our task-oriented considerateness speakers are more likely to groups reinforce the picture of simultaneous call the overlap style competitive and domi- status differentiation and conflict. We find nant, although stimulating.) If such speech that supportive interruptions show different styles are correlated with sex (and participa- patterns from the negative and neutral inter- tion inequalities suggest that they might be), ruptions where conflict dynamics are most members of mixed-sex groups may find likely to be played out. These supportive themselves misinterpreting and reacting against interruptions are the most influenced by group the speech of others. sex composition. In all-male groups we find In summary, we find considerable gender that male-male interruptions are most likely to inequality in the pattern of interruptions, but be supportive. Males interrupting other males it is not easily described by a straightforward in such groups often clarify and even statusldominance process as implied by the encourage the ideas suggested by the inter- expectations states tradition and by most rupted speaker. Although the speech is conversational analysts. Men's actions could disrupted, the group discussion of the topic is be explained by such theoretical traditions, not. Such supportive, topic-continuing inter- but we find women's interruption attempts AMERICAN SOCIOLOGICAL REVIEW and patterns of yielding the floor do not Themes of Male, Female and Mixed Groups." support an image of females simply submit- Small Group Behavior 7:7-18. ting to the higher-status (in expectation states' Berger, Joseph, Susan J. Rosenholtz, and Moms terms, higher task expectation) males. Women' Zelditch, Jr. 1980. "Status Organizing Pro- direct and accept interruptions in a way that cesses." Pp. 479-508 in Annual Review of Sociology. Vol. 6, edited by Alex Inkeles. Palo does not differentiate systematically between Alto, CA: Annual Reviews. males and females. In addition, we find Dabbs, James M., R. Barry Ruback, and M.S. interesting effects of interruptive content and Evans. 1985. "Patterns of Sound and Silence in group sex composition on conversational Group Conversation." In Of Speech and Time: process. These factors suggest that issues Temporal Speech Patterns in Interpersonal involving conversational style, coordination Contexts, edited by A.W. Siegman and S. of speech, and salience of gender identity Feldstein. Hillsdale, NJ: Erlbaum. within groups of varying composition are Drass, Kriss A. 1986. "The Effect of Gender fruitful areas for further research. Identity on Conversation." Social Psychology Quarterly 49:294-301. Eakins, Barbara and Gene Eakins. 1976. "Verbal LYNN SMITH-LOVIN is Associate Profes- Turn-Taking and Exchanges in Faculty Di- sor of Sociology at Cornell University. Her alogue." Pp. 53-62 in The Sociology of the research focuses on the interrelationship of Languages of American Women, edited by Betty affect, identity, emotion, and social behavior. Lou Dubois and Isabel Crouch. San Antonio, She is currently engaged in experimental tests TX: Trinity University Press. of affect control theory. Other projects Fienberg, Stephen E. 1980. The Analysis of involve the role of affective dynamics in Cross-classified Categorical Data. 2nd ed. group discussions and the networks created Cambridge, MA: MIT Press. by organizational affiliation. She is coeditor Fisek, M. Harnit. 1974. "A Model for the Evolution of Status Structures in Task-Oriented of Analyzing Social Interaction (1988), a Discussion Groups." Pp. 53-84 in Expectation book reporting a series of studies of impres- States Theory, edited by Joseph Berger, Thomas sion change and identity maintenance in the L. Conner, and M. Hamit Fisek. Washington, affect control paradigm, and has recently DC: Winthrop. edited a special issue of Social Psychology Heiss, Jerold S. 1962. "Degree of Intimacy and Quarterly on emotions research. Recent Male-Female Interaction." Sociometry 25: 197- works include "Affect, Sentiment and Emo- 208. tion" (Social Psychology Quarterly, March Horvath, William J. 1965. "A Mathematical 1989) and "Emotion as the Confirmation and Model of Participation in Small Group Discus- Disconfirmation of Identity" (Research Agen- sion." Behavioral Science 10: 164-66. das in Emotions, T. D. Kemper, ed., Kanter, Rosabeth Moss. 1977. "Some Effects of Proportions on Group Life: Skewed Sex Rations forthcoming). and Responses to Token Women." American Journal of Sociology 82:965-90. CHARLES J. BRODY is assistant Professor Kollock, Peter, Phillip Blumstein, and Pepper of Sociology at Tulane University. His Schwartz. 1985. "Sex and Power in Interac- current research involves applications of tion." American Sociological Review 50:34-47. latent structures and trends in attitudes Lincoln, James R. and Jon Miller. 1979. "Work concerning abortion and civil liberties. Recent and Friendship Ties in Organizations: A Com- publications have been in the American parative Analysis of Relational Networks. " Administrative Science Quarterly 25: 181-99. Journal of Sociology and Social Forces. Lockheed, Marlaine E. 1985. "Sex and Social Influence: A Meta-analysis Guided by Theory." REFERENCES Pp. 406-29 in Status, Rewards and Influence, edited by Joseph Berger and Morris Zelditch, Jr. Alexander, Victoria D., and Peggy A. Thoits. San Francisco: Jossey-Bass. 1985. "Token Achievement: An Examination of Piliavin, Jane Allyn and Rachael Rosemann Proportional Representation and Performance Martin. 1978. "The Effects of Sex Composition Outcomes." Social Forces 64:332-40. of Groups on Style of Interaction." Sex Roles Anderson, Lynn R. and P. Nick Blanchard. 1982. 4:281-96. "Sex Differences in Task and Social-Emotional Ridgeway, Cecilia L. 1981. "Nonconformity, Behavior. " Basic and Applied Psychology Competence and Influence in Groups: A Test of 3: 109-40. Two Theories." American Sociological Review Aries, Elizabeth. 1976. "Interaction Patterns and 46:333-47. INTERRUPTIONS IN GROUP DISCUSSIONS 435 . 1982. "Status in Groups: The Importance . 1987. "Repetition in Conversation as of Motivation." American Sociological Review Spontaneous Formulaicity. " Text 7:215-43. 47:76-88. Taps, Judith and Patricia Yancey Martin. 1988. . 1983. The Dynamics of Small Groups. "Gender Composition and Attributional Ac- New York: St. Martin's. counts: Impact on Women's Influence and Ridgeway, Cecilia and Joseph Berger, 1986. Likeability in Task Groups." Paper presented at "Expectations, Legitimation, and Dominance the Southern Sociological Society meetings, Behavior in Task Groups." American Sociolog- March I9883 TN. ical Review 5 1:603-17. Taylor, Shelley E., Susan T. Fiske, N.L. Etcoff, Sacks, Harvey, Emanuel Schegloff, and Gail and A.J. Rudeman. 1978. and Jefferson. 1974. "A Simple Systematics for Bases and Organization of Tum-Taking for Conversation." Journal of Personality and So- Language 50:696-735. cial Psychology 36:778-93. Shaw, Marvin E. and Orin W. Sadler. 1965. Thome* Barriey Cheris Kramarae* and "Interaction Patterns in Heterosexual Dyads Henley (eds.). 1983. Gender and Varying in Degrees of Intimacy." Journal of Society. Rowley, MA: Newbury House. Social Psychology 66:345-51. Vuchinich, Samuel. 1984. "Sequencing and Social Structure in Family Conflict." Smith-Lovin, Lynn, John V. Skvoretz, and Social Psycholog- ical Quarterly 47:217-34. Charlotte G. Hudson. 1986. "Status and Partic- Wagner, David G., Rebecca S. Ford, and Thomas ipation in Six-Person Groups: A Test of W. Ford, 1986. ''Can Gender Inequalities Be Skvoretz' Comparative Status Model." Social Reduced?,, American Sociological Review Forces 64:99-1005. 51:47-61. South, Scott J., William T. Markham, Charles M. West, Candace. 1979, "Against Our Will: Male Bonjean, and Judy Corder. 1987. "Sex Differ- Interruptions in Cross-Sex Conversations~w4. ences in Support for Organizational Advance- 81-97 in Language, Sex and Gender. edited by ment. " Work and Occupations 14:261-85. Judith Orsanu, Mariam K. Slater, and Leonore Spang1er3 A. and Loeb Adler. (Annals of the New York Academy Pipkin. 1978. "Token Women: An Empirical of Sciences, Val. 327). Test of Kanter's Hypothesis." American Journal . 1984. "When the Doctor Is a 'Lady': of Sociology 84: 160-70. Power, Status and Gender in Physician-Patient Strodbeck, Fred L., Rita M. James, and Charles Exchanges." Symbolic Interaction 7:87-106. Hawkins. 1957. "Social Status in Jury Deliber- west, candace and D~~ H. zimmerman. 1977. ations. " American Sociological Review u~omen*splace in ~~~~d~~ ~~ll<:~~fl~~ti~~~ 22:713-19. on Parent-Child Interaction." Social Problems Tannen, Deborah. 1980. "Toward a Theory of 24:521-29. Conversational Style: The Machine-Gun Ques- Wharton, Amy S. and James N. Baron. 1987. "So tion." S~~i~ling~i~ti~~Working Paper NO. 73. Happy Together? The Impact of Gender Segre- Austin, TX: Southwest Educational Develop- gation on Men at Work." American Sociological ment Laboratory. Review 52574-87. - 1983. "When Is an Overlap Not an Zimmerman, Don H. and Candace West. 1975. Interruption? One Component of Conversational "Sex Roles, Interruptions and Silences in Style." 4. 119-29 in The First Delaware Conversations." 4. 105-23 in Language and Symposium on Language Studies, edited by Sex: Difference and Dominance, edited by Robert J. DiPietro, William Frawley, and Alfred Barrie Thorne and Nancy Henley. Rowley, MA: Wedel. Newark: University of Delaware Press. Newbury House.