Running Head: EVEN- AND ODD-SIZED GROUPS Getting Even vs. Being the Odd One Out: Conflict and Cohesion in Even- and Odd-Sized Groups Tanya Menon And Kath Philips Electronic copy available at: http://ssrn.com/abstract=1298497 Even- and odd-sized groups Abstract Contrary to people’s intuitive theories about even and odd numbers and groups, this paper argues that odd-sized groups are often more harmonious than even-sized groups. Study 1 found that people view even numbers more favorably than odd numbers and predict that even-sized groups are more peaceful than odd-sized groups. However, Study 2 found that three- and four-person groups without conflict did not differ, but three-person groups with coalitions (two vs. one) produced more positive relationships than four-person groups with coalitions (both two vs. two and three vs. one). Finally, Study 3 involved a natural experiment at Harvard University, and found that White freshmen assigned to odd-sized rooming groups (three or five persons) maintained relationships with White roommates more than did White freshman assigned to even-sized rooming groups (four or six persons), but these patterns did not emerge when Whites roomed with Asians and Blacks. We suggest that a group’s even or odd size is an important situational variable that affects its coalitional structure, conflict management, and cohesion. Electronic copy available at: http://ssrn.com/abstract=1298497 1 Even- and odd-sized groups Everyone knows that two heads are better than one. We also know that two’s company and three’s a crowd. We try our best to remain “even keeled” when we are “at odds” with someone. We seek friends and lovers with whom we are “two peas in a pod,” and we laugh at the “odd couples” we encounter. Although even and odd are simply mathematical concepts, these common idioms reveal that even numbers carry positive connotations and odd numbers carry negative connotations (Hines, 1990; Nishiyama, 2006). Their very definitions reflect these undertones: The word even is synonymous with “balanced, stable, placid, and calm” (even, n.d.), whereas the word odd means “awkward, irregular, peculiar, strange, bizarre, eccentric, and incomplete” (odd, n.d.). Given that people associate even numbers with harmony and odd numbers with conflict, it seems intuitive that social interactions that involve even- or odd-sized groups should reflect these patterns too, and that even-sized groups should experience more harmony than odd-sized groups. In contrast to conventional wisdom that two’s company and three’s a crowd, this paper argues that very often, three’s company, and four’s a crowd. We explore how even- and odd-sized small groups differ and propose that odd-sized groups are often more harmonious than even-sized groups. Our key premise is that even or odd size often affects (a) whether a small group can harness majority influence and (b) how it chooses to use that power. Specifically, odd-sized groups often fracture into a majority and a minority, which enables the majority to influence dissenters (Asch, 1951; Cialdini, 2001; Hastie & Kameda, 2005; Wittenbaum, Merry, & Stasser, 1996). By contrast, even-sized groups experience tension because they often divide along balanced lines (e.g., two versus two in four-person groups; Murnighan, 1978; Shears, 1967), which prevents them from applying majority influence to resolve conflicts (Frank & Anderson, 1971). Majority influence should enable odd-sized groups to cohere more than even-sized groups. We suggest that this is especially true when group members are socially or demographically homogeneous, because people tend to conform within homogenous (versus diverse) groups (Abrams, Wetherell, Cochrane, Hogg, & Turner, 1990; Phillips & Loyd, 2006). Second, we posit that coalitional structure shapes how groups use majority influence. Specifically, majorities within odd and even groups may differ in how they apply their influence. Unlike even groups, in odd-sized groups, a single-person can determine the majority (e.g. two people versus one or three people versus two). When a single individual’s opinion can change the group’s balance, the majority must use influence subtly. However, when the majority possesses a more comfortable margin (e.g. three versus one or four versus two in even groups), they silence dissenters more aggressively. Electronic copy available at: http://ssrn.com/abstract=1298497 2 Even- and odd-sized groups Even and Odd: Intuitive Predictions Versus Real Interaction Intuitive theories about even and odd numbers and groups are important because they can affect how people form groups and how they interact within them. Next, we examine people’s actual behavior within even- and odd-sized groups, and propose that these experiences diverge from people’s predictions. Intuitive Theories about Even and Odd Intuitive theories about even- and odd-sized groups perhaps derive from preferences for and greater perceptual fluency with even rather than odd numbers (Hines, 1990). In English, the word even is a linguistically “unmarked” adjective, i.e., it is the default. Thus, people use it more frequently than the word odd (100 per million to 38 per million; Greenberg, 1966; Thorndike & Lorge, as cited in Hines, 1990). Further, negative prefixes attach to the word even (e.g. uneven) but not the word odd (unodd does not make sense; Hines, 1990; Zimmer, 1964). In addition to greater linguistic and cognitive accessibility of even numbers, at least in the West, people may view even numbers more favorably than odd numbers because they convey symmetry and balance (Nishiyama, 2006).1 Both academic and lay theories associate balance with positive social relationships. Just as early political theorists equated the “balance of power” with stability in foreign policy (Hume, 1752/1985), American government employs “checks and balances” to prevent a tyrannical autocrat from dominating the system (Montesquieu, 1748/1949). Likewise, ordinary people often seek intrapersonal balance (i.e., consistent cognitions without dissonance; Festinger, 1957) and interpersonal balance (Heider, 1958) and symmetry (i.e., having congruent social relationships; Newcomb, 1953). Given a linguistic and cultural heritage that treats evenness preferentially, an intellectual heritage that equates balance with stability and peace, and innate psychological preferences for balanced cognitions and relationships, we hypothesize that Americans will view even numbers more favorably than odd numbers, and that they will expect even-sized groups to exhibit more positive dynamics than odd-sized groups. Relationships in Even- and Odd-Sized Groups People might predict that even-sized groups experience more positive relationships than odd-sized groups, but are these expectations accurate? To answer this question, we draw from classic social psychological research, which has frequently examined group size (Cummings, Huber, & Arendt, 1974). Although group size is “perhaps the least ambiguous and most precisely measurable characteristic of small groups” (O’Dell, 1968: p. 75), critics have characterized group size research as unsystematic (Hackman & Vidmar, 1970; Thomas & Fink, 1963) and yielding 3 Even- and odd-sized groups little consensus (Frank & Anderson, 1971, p. 135). For instance, whereas some researchers celebrated six-person groups because they fostered good communication (Ziller, 1957), inspired solidarity (Bales & Borgatta, 1955), and promoted effective leadership (Bass & Norton, 1951); others noted that five-person groups were neither too small nor too large (Slater, 1958), and that four- and five-person groups experienced high member satisfaction (Hackman & Vidmar, 1970). On the other hand, researchers found that dyads were tense (Bales & Borgatta, 1955; Hackman & Vidmar, 1970; O’Dell, 1968), and that four-person groups experienced conflict (Bales & Borgatta, 1955). In the hundreds of group size studies, although we did not locate a single study that featured the even/odd effect as the focal topic, a few investigators noticed that even- and odd-sized groups differed. For instance, Frank and Anderson (1971) found that people in odd-sized groups (three or five persons) considered their interactions more “warm, pleasant, cooperative, and not serious…and less antagonistic” (p. 147) than those within even-sized groups (two or eight members). They argued that odd groups resolve conflict because they split into a majority and minority, whereas even groups stalemate. Similarly, Bales and Borgatta’s (1955) study of two- to seven-person groups suggested that odd-sized groups function more smoothly, although their sample only contained four groups of each size, and their design involved various confounds and statistical errors (see O’Dell, 1968, for a critique). Finally, Maier (1972) found, contrary to his initial predictions, that triads were more harmonious than four-person groups. These studies offered tantalizing evidence that even- and odd-sized groups might differ. Despite this initial interest in group size, current research shows little interest in the topic. Indeed, group size is so irrelevant that various paradigms exclusively employ a single group size (often three-person groups) for methodological convenience (see Murnighan, 1978 and Phillips, Mannix, Neale, & Gruenfeld, 2004 for a discussion). When we surveyed dissertations over the past 20 years, we found only one small-groups dissertation that featured the term group size in its title.2 One reason that group size research has declined is that, as in much small-groups research, empirical results have outpaced theory-building (Levine & Moreland, 1998). For instance, because prior researchers noted even/odd differences tangentially, they posited a mechanism on a post hoc basis (majority influence), and did not examine it in depth. The current paper further examines the questions of who is subject to majority influence, how it is exercised, and how coalitional structure shapes its usage. Stability from Imbalance in Dyads and Triads 4 Even- and odd-sized groups George Simmel’s comparisons between dyads and triads revealed theoretical processes that could account for the fundamental instability of even-sized groups. Individuals in a dyad are: Confronted only by one another and not by a collectivity…[The triad is] the simplest structure in which the group as a whole can achieve domination over its component members; …the triad can impose its will upon one member through the formation of a coalition between the two others. It is a distinct sociological entity (Coser, 1971: p. 186-187). Thus, although social psychology often maligns the majority tendency to dominate minorities (Asch, 1951; see Peterson & Nemeth, 1996 for an exception), one potentially beneficial consequence is that majority influence enables groups to resolve disputes and achieve coherence. In addition to majority influence, mediation is another mechanism that enables a triad to cohere even if two parties clash (Belliveau & Stolte, 1977; Burt, 1992; Carnevale & Pruitt, 1992; Conlon, Carnevale, & Murnighan, 1994). While a dyad disintegrates if one person leaves it, a mediator, as Simmel (1950) noted, produces “psychological unity which resides in group life… he can carry and direct the very movement of the whole group if the antagonism of the other two tends to paralyze their forces” (p.152).3 Through majority influence and mediation, imbalance moves triads forward, whereas balance paralyzes dyads. Beyond the Triad: Odd versus Even Groups This research moves from dyads and triads to even- and odd-sized groups more generally, and hypothesizes that whereas balance destabilizes even-sized groups, majority influence stabilizes odd-sized groups and encourages them to cohere. Moreover, majorities within odd-sized groups enact their power in a more benign and tolerant way. Odd groups often contain narrow majorities (e.g., two versus one or three versus two) which imply that coalitions can be fluid. If a single individual changes allegiances, the majority can easily become the minority. Given their tenuous status, majorities must be sensitive to the minority’s interests and opinions. Because the majority in odd-sized groups gives more consideration to minority viewpoints, we hypothesize that this increases cohesiveness. Overview To test the hypothesis that odd-sized groups possess several features that create greater stability and cohesion than even-sized groups, we designed three studies that involved multiple methods (survey, laboratory experiment, and natural experiment); groups in various contexts (MBA and undergraduate discussion groups, undergraduate roommates); and different tasks (short term discussion groups, long term living arrangements over a year). 5 Even- and odd-sized groups Studies 2 and 3 moved from people’s predictions about groups to their actual experiences in groups. Study 2 compared three- and four-person groups and manipulated their coalition structure. We did not expect differences to emerge between 3- and 4- person groups when no one disagreed. However, we predicted that the coalitions within four- person groups would impair cohesion more than those within three-person groups. Balanced 2-2 coalitions would lack access to a simple conflict resolution strategy (majority influence). Additionally, odd-sized groups enable narrow majorities (two versus one, three versus two) which encourages the majority to influence without dominating minorities (as in 3-1 groups). Finally, Study 3 considered a natural experiment at Harvard University, where the administration assigns first-year students to rooming groups of one to six persons. We examined how initial freshman group size affected a pair’s likelihood of staying together in their second year. We predicted that Whites assigned to odd-sized rooming groups (three or five persons) would maintain their relationships with their White roommates more than Whites assigned to even-sized rooming groups (four or six persons). Additionally, we predicted that these patterns would be mitigated among diverse pairs (Whites with Asian and Black roommates) because majority influence declines in diverse versus than homogeneous settings (Abrams, Wetherell, Cochrane, Hogg, & Turner, 1990; Phillips & Loyd, 2006). Key Research Assumptions We next identify our key assumptions. In addition to revealing crucial moderators, the conditions described here highlight the challenges of developing a general even/odd theory. Small Groups. Given that small and large groups differ in their structure and behavior (Levine & Moreland, 1998; Simmel, 1950), this research focuses on small groups only. For instance, a majority’s presence can be less salient in a large group (e.g. 25 vs. 26 people), and more complex coalitions can emerge, which might mitigate odd-even effects. Equal Status. We assumed that group members have roughly equal status. If certain people hold leadership positions, veto power (Murnighan & Roth, 1980; Murnighan & Szwajkowski , 1979), or status based on expertise, personality, tenure, or attractiveness (Ridgeway, 1991), they exert disproportionate influence on group decision making (Thomas-Hunt & Phillips, 2003), and can impose conformity without majority influence, so that odd-even size becomes less relevant. Issue Structure. Just as political systems often fracture into two parties (Duverger, 1972), people in small groups also often split into two camps. When issues are dichotomous, odd-sized groups always fracture into a majority and minority. When issues are multi-sided, even/odd dynamics are more complex than the processes we have considered here. Similarly, some groups 6 Even- and odd-sized groups use decision making rules other than a majority rule (e.g. a 2/3 approval). These variations in issue structure and decision making procedure may shift the relevant fault lines (Lau & Murnighan, 2005), such that even/odd size ceases to tap pivotal dynamics. Function. Although we examined groups in various settings (freshman roommates, decision making groups), we did not directly examine whether group function moderates odd-even effects. Given that people may hold different expectations about conformity and conflict in social- and task-focused groups, even/odd effects might vary based on group purpose. Study 1. Intuitive Theories About Odd and Even Numbers We first examined people’s intuitive theories about even and odd numbers and groups. We predicted that people view even numbers more favorably than odd ones, and further, extend these impressions to groups. Method Participants We surveyed 34 American students from the University of Chicago (55.8% male, 39.5% female, and 2 participants failed to report their gender). Their average age was 20.74 years. Procedure Participants viewed even and odd number sequences and rated their first impressions using a 7-point scale (1 = negative, 4 = neutral, 7 = positive). We counterbalanced these sequences such that some participants saw the odd sequence first and others saw the even sequence first. The instructions and number sequences were as follows: Please list your first impressions about this sequence of numbers: 2 4 6 8 10 12 14 16 Please list your first impressions about this sequence of numbers: 3 5 7 9 11 13 15 17 Next, participants read the following vignette: At work, you are assembling a committee to make an important forecast about the company’s strategic position. Obviously, the committee can either be composed of an even (4, 6, 8) or odd (3, 5, 7) number of individuals. We’d like to ask you a few questions about your predictions about being in a group with an even or odd number of people. 7 Even- and odd-sized groups We asked participants whether they believed that even- and odd-sized groups function a) similarly or b) differently (they circled either a or b). They then read a list of group characteristics: peaceful, argumentative, experience positive relationships, resolve conflict productively, develop coalitions, result in exclusion, and use the majority rule to resolve conflicts. Finally, we asked whether they preferred to participate within even-sized or odd-sized groups. They circled either the word even or the word odd to indicate their choices. Results Participants preferred even numbers (M = 4.85, SD = .99) to odd numbers (M = 3.97, SD = 1.14), F(1, 32) = 7.30, p = .01. People’s preferences for even numbers extended to even-sized groups. Among the participants, 85.3% expected even- and odd-sized groups to function differently, t(33) = -5.73, p < .01. Specifically, 73.5% assumed that even-sized groups had more positive relationships, t(33) = 3.06, p < .01, and 67.6% assumed that they were more peaceful, t(67) = 2.17, p < .05. On the other hand, 67.6% expected odd-sized groups to experience more conflict, t(33) = -2.17, p < .05; 76.5% assumed that they involved more exclusion, t(33) = -3.56, p < .01; and 88.2% assumed that they used the majority rule more, t(33) = -6.82, p < .01. With regard to coalitions, 73.5% expected that coalitions were more prevalent within even-sized groups, t(33) = 3.06, p < .01. People preferred to participate within even-sized and odd-sized groups equally (55.9% preferred odd-sized groups), t(33) = -.68, p = .23. Discussion Study 1 assessed people’s intuitive theories about even and odd numbers. As predicted, people favored even sequences over odd sequences. Further, people assumed that even-sized groups were more peaceful and positive and experienced less conflict and exclusion than odd-sized groups. On the other hand, people preferred even- and odd-sized groups equally, perhaps because of the group’s tasks (i.e., it was a decision making group in which quick, efficient decisions would be favorable). These intuitive preferences for even groups may arise from culturally-derived positive associations with evenness, rather than from their analysis of social relationships. In the next study, we examine how people actually experience social relationships within odd and even groups. Study 2: Manipulating Coalitions in Three- and Four-Person Groups To simplify the empirical context, Study 2 examines only three- and four-person groups. We seek to directly test the conflict resolution mechanisms that people within even and odd-sized groups use. First, we expect no differences between three- and four-person groups when they 8 Even- and odd-sized groups experience no conflict, and thus suggest that even/odd effects derive, not from group size per se, but from coalitional structure and conflict resolution processes. Additionally, this study compares unbalanced three-person coalitions (two versus one) with both balanced (two versus two) and unbalanced (three versus one) four-person coalitions. We argue that regardless of whether four-person groups experience a balanced or unbalanced distribution of opinions, they are likely to experience the group as more conflict-ridden and less cohesive than three-person groups. When even groups split into balanced forces, they face unresolvable conflict and instability. Furthermore, we predicted that moderately unbalanced groups (two vs one) would use subtle social pressure and that extremely unbalanced groups (three vs one) would aggressively silence dissenters. We experimentally manipulated the group’s coalitional structure based on participants’ initial opinions about a discussion topic. Thus, we created five conditions: (a) three-person group, members all agree; (b) three-person group, two people agree and one disagrees; (c) four-person group, members all agree; (d) four-person group, three people agree and one disagrees; and (e) four-person group, two agree and two disagree. In sum, our theory relies on coalitional structure, so we did not expect three- and four-person groups to differ when all members agreed. We hypothesized that three-person groups (two vs. one coalitions) would evoke more positive relationships than four-person groups both when four-person groups were balanced (two vs. two) and when they were extremely unbalanced (three vs. one). Method Participants The participants included 189 students from Northwestern University (50% male). Of these, 94 were MBA students who participated in a class exercise, and 95 were undergraduates who participated in the lab and were paid $10. We excluded 14 MBA student participants because they failed to complete the pretest or because of experimenter error in creating group assignments. We pooled these data because participants in the groups did not vary. Procedure All participants read the Carter Racing case (Brittain & Sitkin, 1988), in which they assumed the role of an automobile racing team deciding whether to race (and potentially win a large endorsement) or not race (because of the risk of a potentially disastrous engine failure). Although participants tend to select the risky option (racing), the correct choice, based on the data, is to not race. (Indeed, the failure rates come from the data that NASA scientists viewed before they decided to launch the doomed Challenger space shuttle.) Participants made a race/no race 9 Even- and odd-sized groups decision and then answered a brief pretest which assessed their confidence about their decision using the following questions: How confident are you about your decision to race or not to race? This question was scaled from 0 percent to 100 percent in 10 percentage increments (0=0% not at all confident, 11=100% completely confident) Assume you have $250 at your disposal, how much of it would you be willing to bet that you have made the right decision? This question response was scaled from $0 to $250 in $25 increments (0= $0, 11=$250) Do you expect the other members of the class to agree with your decision? (1= not at all to 7=very much) Do you feel you have all the information you need to make this decision? (1=not at all, 7=very much) Next, we randomly assigned participants into three- and four-person groups that varied in their coalition structure. As mentioned before, our design contained five alternative coalitional structures: (a) three-person group, members all agree (n = 12); (b) three-person group, two people agree and one disagrees (n = 11); (c) four-person group, members all agree (n = 8); (d) four-person group, three people agree and one disagrees (n = 9); and (e) four-person group, two agree and two disagree (n = 8). Given that most participants in this exercise elect to race, we always assigned nonracers to the minority. After we assigned the participants to groups and before they discussed the case, participants completed another survey in which they rated how well they knew each group member, rated how close they felt towards each member, and predicted their group’s dynamics. They predicted the extent to which each of the following outcomes was likely (All items used a 9point scale in which 1 = not at all and 9 = completely): To be difficult to resolve disagreements with the other people with whom you have been assigned to discuss this situation. That some of the people you have been assigned to work with have difficult personalities. To be difficult to coordinate interactions in a group of this size. Finally, participants discussed the decision for 20 minutes within their small groups, submitted a group decision to the experimenter, and then completed a final survey. Varimax factor analysis revealed two factors that tapped group process: Group cohesion (Alpha = .88). The items included were as follows: We were a cohesive group; We felt comfortable working together; I felt strong ties with this group; I identified with this group; and I feel good about the group. 10 Even- and odd-sized groups Satisfaction with group decision making (Alpha = .72). The items included were as follows: I am satisfied with our decision-making process; The majority of people in the group agreed with my view; We had enough time to discuss the case; I am satisfied with our final decision. Next we asked participants about their groups’ conflict-resolution techniques. We asked participants whether the group was characterized by each of the following: Mediators/leaders (Alpha = .64). The items included were as follows: One person tended to play the role of a mediator to resolve our disagreements; One person acted as a leader who facilitated our discussion. Majority rule: The item was worded as follows: We used the majority rule to solve our disagreements. Feeling excluded (alpha=.74). The items included were as follows: I sometimes felt excluded; I felt like a useless group member; Some of the people I was assigned to work with had difficult personalities; I often simply went along with the others even if I didn’t agree with them. Avoiding exclusion. The item was worded as follows: I felt badly if one person seemed excluded from the group. Finally we debriefed all participants and paid the undergraduates. Results Given the multi-level data, (individuals who were nested within discussion groups), we used hierarchical linear modeling (HLM). We included individual-level variables (e.g. group process ratings), group level characteristics (e.g. group size, coalition structure), and interactions between individual and group characteristics (for examples and discussions of similar analyses, see Kreft & de Leeuw, 1998; Wilk & Moynihan, 2005). Differences Between Racers and Nonracers Among the participants, 137 elected to race and only 39 chose not to race. Racers and nonracers did not vary in their initial confidence, F(1, 141) = 1.45, n.s.; in the amount they bet that their decision was accurate, F(1, 142) = .15, n.s.; or in whether they expected that others would agree with them, F(1, 142) = 1.11, n.s. However, nonracers were more likely than racers to believe that they had all the information necessary to make an accurate decision, F(1, 142) = 7.50, p < .01 (See Table 1a). Differences Between Three- and Four-Person Groups Next, we examined participants’ responses after we assigned them to groups but prior to their actual interaction within those groups. Using HLM to control for the group, we found that participants who expected to interact in three-person groups did not differ from those who expected 11 Even- and odd-sized groups to interact in four-person groups. They did not differ in how well they knew their fellow group members, B = -.15, t(52) = -.72, n.s.; or in how close they felt to them, B = .04, t(51) = .08, n.s. (See Table 1b).. Additionally, in contrast to Study 1 in which we highlighted a group’s oddness versus evenness, participants we assigned to three- and four-person groups expected their groups to operate similarly. Thus, participants we assigned to three- and four-person groups expected disagreements to be equally difficult to resolve, B some people to have difficult personalities, B = .24, t(52) = .90, n.s.; they equally expected = .14, t(52) = .55, n.s.; and they equally expected that interactions would be difficult to coordinate in a group of their particular size, B = .26, t(52) = -1.33, n.s. Differences Following Group Discussion in Three- and Four-Person Groups Finally, we examined outcomes following the group discussion and introduce coalition structure as a manipulated factor. As predicted, the group’s coalitional structure shaped its subsequent cohesion (See Table 1c). Specifically, participants in both two vs. two and three vs. one groups reported lower cohesion than participants in two vs. one groups (two vs. two: B = -.74, t(43) = -2.44, p < .05; three vs. one: B = -1.12, t(43) = -.3.60, p <= .01), but they did not differ from one another B = -.38, t(43) = -1.05, n.s. Likewise, compared to participants in two vs. one coalitions, participants in four-person groups rated their group processes more negatively (three vs. one: B = -1.00, t(43) = -2.37, p <= .05; two vs. two: B = -1.15, t(43) = -2.65, p = .01). Three vs. one and two vs. two groups did not differ from one another: B = .15, t(43) = .35, n.s. One alternative explanation for these results is that they are simply due to group size. Perhaps four-person groups were more difficult to manage than three-person groups. However, four vs. none and three vs. none groups did not differ from one another with respect to cohesion, B = -.15, t (43) = -.41, n.s., or from two vs. one groups (four vs. none: B = -.41, t (43) = -1.36, n.s.; three vs. none: B = -.56, t(43) = -1.66, n.s.) and satisfaction with group process, (three vs. none: B = .72, t(43) = 1.77, p < .10, and four vs. none: B = .25, t(43) = .59, n.s.). This suggests that coalition structure within these groups, rather than simply group size, mattered. A second possibility is that disgruntled minority participants drove these findings. Therefore, we excluded minorities and examined racers only. Even the racers in the two vs. two and three vs. one groups identified with the groups less (three vs. one: < .01; two vs. two: B B = -1.21, t(43) = -3.96, p = -.74, t(43) = -2.79, p < .01) and were more dissatisfied with the group’s 12 Even- and odd-sized groups decision making process (three vs. one: B = -.78, t(43) = -1.98, p = .05; two vs. two: B 13 = -1.48, t(43) = -3.49, p < .01) than majority racers in two vs. one groups. Another indicator of group cohesion is whether participants’ individual opinions conformed to the group’s decision following the discussion. We coded whether a group was united (i.e., if each individual’s private opinion conformed to the group’s opinion) or whether it contained private dissenters (i.e., individuals who remained unpersuaded by the group’s opinion). As expected, there was greater unity in two vs. one groups (nine united groups vs. two with private dissenters), compared to two vs. two groups (three united groups vs. five with private dissenters), χ 2(1, N = 19) = 3.91, p < .05; and three vs. one groups (two united groups vs. seven with private dissenters), χ 2 (1, N = 20) = 7.10, p < .01. Note that our hypotheses did not predict that three-person groups would be more accurate than four-person groups. Thus, although two vs. one groups reported greater conformity and satisfaction, they did not differ from three vs. one and two vs. two groups in their racing decision, χ 2 (2, N = 28) = 3.24, n.s. Dispute Resolution Mechanisms We next assessed whether group processes varied within three- and four-person groups. Two vs. one and three vs. one groups used a leader/mediator to resolve disputes equally, (three vs. one: B = .05, t(43) = .09, n.s.; two vs. two: B = .32, t(43) = .57, n.s.). However, participants in three vs. one groups used the majority rule more than did two vs. one participants, B = 1.63, t(43) = 2.20, p < .05. Whereas the three vs. one groups explicitly used the majority rule to silence dissenters, people in two vs. one coalitions used majority influence in a more subtle manner (Rico, SanchezManzanares, Gil, & Gibson, 2008). Thus, participants in three vs. one groups felt more excluded than participants in two vs. one groups, B = .69, t(43) = 2.55, p = .01. Compared to nonracers in two vs. one groups, nonracers in three vs. one groups were more dissatisfied with the decision making process (two vs. one: M = 6.65, SD = 1.25; three vs. one: M = 4.86, SD = 2.08, F(1, 17) = 5.31, p < .05) and felt marginally more excluded than dissenters in two vs. one groups (two vs. one: M = 1.55, SD = .62; three vs. one: M = 3.06, SD = 2.25, F(1, 17) = 4.15, p < .10). Because a lone dissenter seems especially unreasonable in a three vs. one group, participants felt less bad about excluding that person, B = -1.61, t(43) = -2.44, p < .05. Discussion Study 2 suggests that an even/odd effect arises from unique coalitional structures, which can shape how conflict emerges and how people resolve it. Given that dispute resolution is Even- and odd-sized groups unnecessary when all participants agree, we did not expect or find that three- and four-person groups differ when people shared a common viewpoint. However, when people disagreed, threeperson groups (two vs. one) reported greater cohesion and more positive relationships than fourperson groups (both three vs. one and two vs. two groups). Participants resolved disputes differently within these coalitional structures. In contrast to the four-person groups, people in two vs. one groups did not silence dissenters via the majority rule or exclusion and instead reported that they felt especially bad if they excluded others. As a result, they gained conformity through subtle (rather than explicit) coordination (Rico et. al, 2008) and maintained cohesion, both publicly and privately. In addition to finding negative effects with respect to balance (two vs. two groups were more hostile than two vs. one groups), we also found that imbalance evoked hostility under specific circumstances (three vs. one groups were also more hostile than two vs. one groups, Mannix, 1993; O’Leary & Mortensen, 2005). Whereas balance leads to coalition formation and polarization (Murnighan, 1978), extreme imbalance (e.g. a single resistant dissenter) also disrupts harmony in an otherwise unanimous group. When facing a large majority, a lone dissenter seems especially wrong and unreasonable. Further, such dissenters undermine the group’s ability to form a powerful unanimous front to external constituents (Pfeffer, 1992). Consider an occasion when the Supreme Court faced an eight vs. one situation and Chief Justice Warren persuaded the lone dissenter: “Stan, you’re all by yourself on this now. You’ve got to decide whether it’s really the best thing for the country” (Schwartz, 1993, p. 298). Thus, within an extremely imbalanced group, an overly aggressive majority might feel especially justified in its right to exclude and undermine minorities. These differences between two vs. one and three vs. one groups highlight the need for more nuanced theories about tokens (e. g., Ely, 1994; Levine & Moreland, 1998) and majority/minority relationships (Levine & Thompson, 1996). Note that participants initially expected that three- and four-person groups would not differ. In contrast to Study 1, which indicates that people’s expectations about even/odd groups differ, Study 2 did not highlight group size or ask about it directly. Further, we surveyed people before they realized that they were in the majority or the minority, so numerical issues were less salient (Kanter, 1977). As such, although a group’s even/odd size crucially shapes group processes, this issue can recede into the background when people intuitively understand groups and how they operate. 14 Even- and odd-sized groups Finally, whereas much research has compared dyads and triads as the building blocks of social relationships (Caplow, 1968; Simmel, 1950), we propose that three- and four-person groups offer a fruitful starting point for thinking about group structure and conflict. Although three- and four-person groups enable us to understand structures and processes within even- and odd-sized groups, a key question is how patterns within these groups generalize to larger groups, and whether an even/odd effect emerges. Additionally, we move from a laboratory experiment to a natural experiment that examines people in real groups. Study 3. Roommates at Harvard University: A Natural Experiment We examined a unique natural experiment at Harvard University in which students made meaningful decisions about their living arrangements. The university administration assigns freshmen to roommate groups that vary in size. In their second year, students select their own roommates. We can thus examine how group size affects cohesion. Further, we can examine how ethnic diversity shapes odd-even effect. Research Context Before incoming Harvard freshmen arrive on campus, they answer questions about their habits (e.g. neatness, studiousness, smoking) and interests (e.g. sports, music). The university uses this information to match students in freshmen dormitory rooms with between one and six persons. The administration attempts to match compatible personalities and also integrates students ethnically. Given that the freshman distribution maximizes diversity, and given that the administration attempts to ensure basic compatibility between roommates, these assignments are not perfectly randomized. However, it is unlikely that students request an odd or even number of roommates. According to the Harvard Dean of Freshman, Thomas Dingman, students rarely report preferences or aversions to roommates based on ethnicity, and the housing assignment process does not accommodate such requests (Dingman, personal communication). In the sophomore year, students choose blocking groups of up to 20 persons and move into one of 13 “houses” where sophomores, juniors, and seniors live together, in contrast to the 16 dorms, which house only freshmen. Students list their top four choices and participate in a lottery to determine whether they receive their preferred house with theor preferred group. Roommate choice indicates interpersonal relationships. We infer that a relationship was relatively positive when students decide to maintain it, because people who choose to stay together ensure that they remain in close contact for the next three years. By contrast, when people decide not to live together in the second year, they tend to live in separate houses, and this reduces their likelihood of dining together or even seeing each other regularly, especially if they live at opposite ends of the 15 Even- and odd-sized groups campus. This makes it reasonable to infer that people who chose not to room together often experienced negative or neutral rather than positive relationships. Creating the Dataset To create the dataset, we used the freshman phone directory and the annual house yearbook entries to arrange students from Harvard’s class of 1995 into their freshman and sophomore rooms. We recorded each freshman roommate pair as a case, and coded whether they remained in the same dorm in the second year or lived in different dorms. We next coded ethnic information (i.e., was the pair White-White; White-Asian; Asian-Black; etc.). We coded ethnic information based upon the students’ appearance in photographs and on their names as reported in the freshman and house yearbooks.4 Limitations Given the indirect route necessary to collect data, limitations arise in two specific areas: the coding of dorm placement and the coding of ethnicity. Specifically, because the university keeps information such as student ethnicity and roommate selection private, we had to make inferences based on the available data. Coding Dorm Rooms We had no direct access to students’ choices. We simply examined whether the students lived in the same or different dorms as sophomores. If they were not in the same second year dorm, we inferred that they chose to separate. However, if they chose to separate, there was a 1 in 13 chance that the housing lottery randomly assigned them to the same dorm. Because such processes are due to chance, they should not affect even- and odd-sized groups systematically. Coding Ethnicity Second, without access to university records, we coded ethnicity using students’ yearbook photos. Thus, if people appeared Black or White in their photographs although they were of mixed descent, they were coded according to their appearance for the purposes of this study. Additionally, given the relatively limited representation of minorities in the sample, our categories lumped many groups together (e. g., Black included American, Caribbean, and African; Asian included American, Korean, Chinese, Japanese, etc.). Finally, we categorized people as mixed or “other” if we couldn’t clearly identify their ethnicity. An independent coder validated the first author’s perceptions of the students' physical appearance, with high inter-rater reliabilities (Whites: 87%; Asians: 97%; and Blacks: 86%). Results 16 Even- and odd-sized groups Given the multi-level data (i.e., individuals are nested within rooming groups), we again used HLM to control for the group-level effects. We included controls for gender, freshman room size, and odd-even size, and included an interaction between odd-even size and whether roommates shared the same or different ethnicity. The dependent variable was dichotomous (do roommate pairs stay or break), so we used binomial logit. We found an interaction between odd-even room size and whether roommates were same or different race, B = -.52, t(1561) = -2.06, p < .05. Specifically, we found an odd-even main effect among White-White pairs, whereby they stayed together more in odd-sized groups (three and five persons) than in even-sized groups (two, four, and six persons; Figure 1 represents this zig-zag pattern). Although White pairs in two- and three-person groups did not differ in their likelihood of staying together, B = -.14, t(108) = -.36, n.s., White pairs in three- to six-person groups differed. White pairs in three-person groups stayed together more than Whites in four-person groups, B = .63, t(210) = -2.42, p < .05; Whites in four-person groups stayed together less often than Whites in five-person groups, B = -.73, t(176) = -2.11, p < .05; and Whites in five-person groups stayed together more than Whites in six-person groups, B = -.70, t(51) = -1.98, p = .05. Gender, B = -.14, t(411) = -.82, n.s., and freshman room size, B = -.82, t(411) = -.82, n.s., were nonsignificant. By contrast, we did not find odd-even effects between Whites and their Black and Asian roommates, B = -.15, t(268) = -.65, n.s. This could occur because people feel uncomfortable both when ingroup members exclude them and when they exclude fellow ingroup members. As a result, whereas people expect ingroup members to conform, they expect minorities (who are already outsiders due to their ethnicity) to disagree with ingroup members (Phillips, 2003; Phillips et. al., 2004). This mitigates majority influence among diverse pairs and reduces odd/even effects.5 Discussion Whereas most groups research employs laboratory experiments (see Moreland, Hogg, & Hains, 1994 for a discussion), this study found that people in real even- and odd-sized groups differed in their relationships. In contrast to intuitive predictions, Whites in odd-sized groups maintained their roommate ties more often. We propose that this occurs because odd-sized groups can readily employ majority influence to resolve conflicts. Consistent with this mechanism, the effect pertains to same-ethnicity roommate pairs. Because congruence implies that similar people should agree and dissimilar people should disagree (Phillips, 2003; Heider, 1958), the majority’s power to influence a dissenting minority declines among diverse people, and even-odd effects disappear. General Discussion 17 Even- and odd-sized groups A common challenge to social relations is that groups polarize along fault lines (Lau & Murnighan, 2005). Civil wars between ethnic subgroups fracture countries such as Iraq and the former Yugoslavia; “Red” and “Blue” states in the United States hold fundamentally incompatible values and lifestyles; and governing institutions such as Congress stall due to political conflict. At a micro level, top management teams (Hambrick, 1994; Peterson et. al., 1998), work teams (Lau & Murnighan, 2005; O’Reilly, Williams, & Barsade, 1998; Pelled, 1996), judges (Sunstein, Schkade, Ellman, & Sawicki, 2006), juries (Davis, Kerr, Atkin, Holt, & Meek, 1975; Guarnaschelli, McKelvey, & Palfrey, 2000; Hastie, Penrod, & Pennington, 1983; Saks, 1977), and families (Caplow, 1968; Grusky, Bonacich, & Webster, 1995) also divide along fault lines. This paper examines how small groups fracture in distinctive ways due to a seemingly inconsequential factor: their even or odd size. As such, this research connects classic studies on group size with recent themes such as diversity (Williams & O’Reilly, 1998) and demographic fault lines (Lau & Murnighan, 2005; Pfeffer, 1983). While early research examined group size without considering coalition structure and diversity as moderators, and recent research treats group size as fundamentally uninteresting and unimportant, the current research asserts that a group’s odd or even size is a crucial situational feature that shapes its decision making, coalition structure, and cohesion. In three studies that used multiple methods, we examined two issues: people’s predictions about even- and odd-sized groups and the actual cohesion within these groups. We found that people held more positive attitudes towards even than odd numbers and further predicted that people in even-sized groups experienced more positive social relationships (Study 1). Studies 2 and 3 examined people’s actual experiences in these groups. Study 2 found that three- and four-person groups did not differ when all parties agreed, but they differed when they fractured into coalitions. Four-person groups (both two vs. two and three vs. one coalitions) experienced more hostile group relationships than three-person groups (two vs. one coalitions). Both balance (e. g., two vs. two coalitions, which cause unresolvable conflict) and extreme imbalance (e.g., three vs. one coalitions, which lead majorities to unilaterally exercise power) undermine positive group relationships (Mannix, 1993; O’Leary & Mortensen, 2005). In a natural experiment at Harvard University (Study 3), first-year students assigned to odd-sized roommate groups stayed with roommates more often than those who were assigned to even-sized roommate groups, and ethnic homogeneity moderated this effect. We argue that these differences do not arise from group size per se but from coalitional structure. In sum, these studies systematically replicate 18 Even- and odd-sized groups even/odd effects that appeared occasionally and tangentially decades ago in the group size literature, and examine their underlying mechanisms. Limitations Although our evidence indicates that even- and odd-sized groups exhibit critical differences, as mentioned previously, a parsimonious even/odd theory may be elusive given that group size (Simmel, 1950), member status (Thomas-Hunt & Phillips, 2002), issue structure, and salient decision rules (Hastie & Kameda, 2005) likely all moderate the effect. Second, this research emphasizes the ways that groups resolve conflict once they fracture, rather than how they tend to fracture in the first place. Thus, we manipulated disagreement in Study 2, and, in Study 3, we assumed that roommate groups experience disagreement when they live in close quarters for one year. For instance, Murnighan (1978) cited several studies indicating that tetrads tend to split in two vs. two groups (Shears, 1967; Willis, 1962) rather than three vs. one coalitions. Further research should consider how triads, tetrads, and larger groups naturally split into particular coalitional structures. Next, although we considered group heterogeneity, we did not directly manipulate social identity and categorization (Tajfel & Turner, 1986). If we prime people to identify with a group, this should heighten even-odd effects; by contrast, if we prime within-group distinctions (race, gender, interests) we expect even/odd effects to weaken. Although Study 2 manipulated heterogeneity through opinions and Study 3 examined ethnic heterogeneity that naturally occurred within groups, further research should experimentally manipulate the group’s salience as an overarching identity and within-group categories (Gaertner & Dovidio, 2000). On a related note, our research has not considered cross-cultural differences. On one hand, odd/even effects might decrease within consensus-building, collectivistic cultures, because their strong norms encourage conflict avoidance and compromise (Morris, et. al., 1998), which imply that collectivistic groups fracture less often, regardless of whether a group is even- or odd-sized. However, once collectivistic groups fracture, even/odd effects can be more pronounced. People who are especially concerned about exclusion (e.g., two vs. one groups) may readily come in line. Further, if groups fracture into evenly matched coalitions (e.g. two vs. two), collectivists might view the rival coalition as an outgroup and avoid them more (Leung, 1987). Implications Expectations Versus Preferences Study 1 reveals that our results for real groups (Studies 2-3) are counterintuitive. People like even numbers and predict that even-sized groups experience more positive relationships. Just 19 Even- and odd-sized groups as marketers tend to sell products in even numbers (e.g. chairs, dishes, eggs, and baked goods), people may compose social groups in even numbers, because they associate them with balance and harmony. Further, people could feel apprehensive when they enter odd-sized groups. For instance, when we presented this research at a conference, one professor observed that her daughter, an entering college freshman, worried because the university had assigned her to a three-person dorm room. Expectations also affect how people explain and resolve conflict. If people in even-sized groups expect smooth relationships, conflict is especially upsetting. By contrast, if people in oddsized groups expect conflict, they may emphasize situational rather than personalistic attributions (“It’s not easy being a threesome”), which make conflict more easily resolvable (Morris, Larrick, & Su, 1999). Finally, people’s expectations affect coalitional strategies. If people assume that balance benefits groups, they might join the weaker powers to create a countervailing force rather than bandwagoning, i.e., joining stronger parties to dominate weaker parties (Walt, 1987). Although people predict that balance leads to positive outcomes, the current research suggests that they may instead prefer interactions within unbalanced coalitions. Shared Cognition Contrary to people’s expectations (Study 1), we found that even-sized (four-person) groups used the majority rule more than odd-sized (three-person) groups (Study 2). Whereas even-sized groups engage in explicit coordination via communication to enforce agreement (e.g., directly invoking the majority rule), odd-sized teams engage in implicit coordination (e.g., subtly inducing conformity through majority influence; Rico et. al., 2008). Because odd-sized groups achieve cohesion through implicit rather than explicit coordination, they may form transactive memory systems more rapidly, i.e., they come to share cognitive processes, such as encoding, storing, and retrieving information (Hollingshead, 1998; Levine, Resnick, & Higgins, 1993; Moreland, Argote, & Krishnan, 1996; Wegner, 1987). On other hand, people in even-sized groups may form transactive memory systems within subgroups. Because they do not acknowledge that rivals possess valuable knowledge (Menon & Pfeffer, 2003; Menon, Thompson, & Choi, 2006), they may replicate that knowledge rather than create shared memory systems Finally, further research might examine the patterns of transactive memory errors that even- and odd-sized groups develop. For instance, people in odd-sized groups may display false consensus (Ross, Greene, & House, 1977), and assume that their peers agree with their viewpoints. 20 Even- and odd-sized groups By contrast, people in even-sized groups may exhibit a “false dissent” effect and assume that rival group members disagree. Designing and Forming Effective Groups Given that group harmony facilitates some tasks (e.g., implementation) whereas debate and disagreement facilitate other tasks (e.g., creativity and brainstorming; Janis, 1982; Peterson & Nemeth, 1996; Sutton & Hargadon, 1996; Swaab, Phillips, Diermeier & Medvec, 2008), further research might compare performance differences between even- and odd-sized groups. Ultimately, such research would enable people to selectively deploy even- or odd-sized groups for appropriate tasks. Further, this simple group characteristic helps people predict the distinct group dysfunctions within even- and odd-sized groups. Specifically, disruptive coalitions within even-sized groups can inhibit cooperation and collaboration (Jehn & Mannix, 2001); while odd-sized groups suffer from conformity and groupthink (Janis, 1982). Perhaps odd-sized groups benefit from different organizational “repairs” (Heath, Larrick, & Klayman, 1998) such as the use of a devil’s advocate (Janis, 1982), while even-sized groups benefit from other strategies (e.g., assigning an expert or leader to help resolve disputes). Finally, in contrast to traditional group interventions such as discussing conflicts directly or enhancing morale through team-building activities, changes in group size offer a relatively simple and subtle means to shift group behavior (Moreland, Levine, Wingert, 1996). For instance, although research has emphasized the challenges that newcomers face (Choi & Levine, 2004; Gruenfeld, Martarona, & Fan, 2000; Levine & Moreland, 1991), adding a newcomer to a deadlocked four-person group also turns it into a five-person group, and encourages conflict resolution through majority influence. Conversely, if a group relies on majority influence too readily, removing or adding a member can alter its numerical context and increase debate. Conclusion This research extends Simmel’s (1950) theories about dyads and triads to tetrads and small groups more generally. Our key point is that sometimes a group essentially operates as a polarized dyad, and other times it coheres as a stable, unified entity. A simple quantitative feature of the group—odd or even size—powerfully shapes which dynamic emerges. Indeed, precisely because no one wants to be the odd one out, people in odd-sized groups may conform, agree, and achieve stability, while people in even-sized groups are often at odds. 21 Even- and odd-sized groups End Notes 1 According to Nishiyama (2006), these preferences arise from cultural factors. The Chinese, for example, hold negative connotations about even numbers, perhaps because yin-yang theories associate odd numbers with masculinity (yang) and even numbers with femininity (yin). In Japanese, “The number 2 means to divide (or to part, separate); 4 is associated with death; and 6, as in the word rokudenashi, means good for nothing” (p. 479). Likewise, a monetary gift of $101 (vs. $100) appears strange to Americans, but Indians traditionally offer monetary gifts in odd sums. Whereas Americans celebrate the “Sweet 16” birthday or a “debutante” ball at age 18, Latin cultures mark the 15th birthday (quinceanera), and Japanese celebrate the 3-5-7 festival to mark a child’s odd birthdays. 2 Three other dissertations included the term group size, but they were not small group research. Two concerned class size in education and one concerned mental institutions. 3 A broker can also be a partisan who connects these parties for personal profit (Burt, 1992; Simmel, 1950) 4 We examined Whites, Asians, and Blacks in this analysis because these were the largest minority groups. Hispanics, East Indians, Middle Easterners, and mixed race participants were excluded because they lacked critical numbers in this dataset. 5 One possibility is that odd-sized groups may be generally more stable than even-sized groups. To test this, we examined whether odd-sized groups were more likely than even-sized groups to stay together in their entirety. We analyzed the data at the group level and cross-tabulated group size and group outcome (entire group stayed together, group disbanded). Two-person groups stayed together in their entirety more than three-person groups (two-person: 57 groups stayed, 71 disbanded; three-person: 30 groups stayed, 75 disbanded), χ 2 (1, N = 233) = 6.28, p = .01. Threeperson groups stayed together more than four-person groups, (four-person: 17 groups stayed, 145 disbanded) χ 2 (1, N = 267) = 14.35, p < .01; and five-person groups stayed together more than did six-person groups, (five-person: 7 groups stayed, 28 disbanded; six-person: 0 groups stayed, 18 disbanded), χ 2 (1, N = 53) = 4.15, p < .05. These effects could simply reflect the unsurprising effects of group size, given that larger groups have more linkages to maintain. However, fiveperson groups were more likely to stay together as a whole than four-person groups, though this trend missed significance, χ 2 (1, N = 197) = 2.43, p = .12. 22 Even- and odd-sized groups Study 2 Means Table 1a: Pretest Characteristics of Racers and Nonracers Pretest Scores Racers Variable Initial confidence Amount of bet Expected agreement Adequate information Nonracers M 8.16 7.32 4.49 SD 1.71 2.54 1.27 M 7.68 7.11 4.21 SD 2.47 3.11 1.17 3.41 1.44 2.57 1.55 Table 1b: Three-person and Four-person Groups Prior to Interaction Group assignment Three-Person Groups Four-Person Groups Variable M SD M SD Relationships to group members Knew group 1.67 1.26 1.52 .86 members well Closeness 3.35 2.22 3.40 2.17 Predicted difficulty of… Resolving group 3.14 1.76 3.38 1.76 disagreements Personalities 2.31 1.66 2.45 1.54 within the group 1.92 1.14 2.19 1.39 Coordinating interactions within a group of their size Table 1c: Comparison of Cohesion Scores by Coalitional Structure (Study 4) Cohesion scores Coalition structure M SD three vs. none 6.78 1.40 two vs. one 7.33 1.10 four vs. none 6.91 1.24 three vs. one 6.21 1.32 two vs. two 6.59 1.27 23 Even- and odd-sized groups Figure 1 Percentage of first year White-White and White-Minority pairs who stayed together in the second year. 60 50 40 30 20 10 0 white-white pairs white-minority pairs 2 person 3 person 4 person 5 person 6 person groups groups groups groups groups 24 Even- and odd-sized groups References Abrams, D., Wetherell, M., Cochrane, S., Hogg, M. A., & Turner, J. C. (1990). Knowing what to think by knowing who you are: Self-categorization and the nature of norm formation, conformity and group polarization. British Journal of Social Psychology, 29(2), 97-119. Asch, S. E. (1951). Effects of group pressure upon the modification and distortion of judgment. In H. Guetzkow (Ed.). Groups, leadership and men. Pittsburgh, PA: Carnegie Press. Bales, R. F., & Borgatta, E. F. (1955). Size of group as a factor in the interaction profile. In P. A. Hare, E. F. Borgatta & R. F. Bales (Eds.), Small group studies in social interaction (pp. 396-413). New York: Alfred A. Knopf. Bass, B. M., & Norton, F. M. (1951). Group size and leaderless discussions. Journal of Applied Psychology, 35(6), 397-400. Belliveau, L. & Stolte, J. F. (1977). The structure of third party intervention. Journal of Social Psychology, 103(2), 243-250. Brittain, J., & Sitkin S. (1988). Carter Racing Case and Teaching Notes. Stanford Case System (#S-OB-24), Graduate School of Business, Stanford University. Burt, R. (1992). Structural holes: The social structure of competition. Cambridge, MA: Harvard University Press. Caplow, T. (1968). Two against one: Coalitions in triads. Oxford, England: Prentice-Hall. Carnevale, P. J. & Pruitt, D.G. (1992). Negotiation and mediation. Annual Review of Psychology, 43, 531-582. Choi, H-S, & Levine, J. M. (2004). Minority influence in work teams: The impact of newcomers. Journal of Experimental Social Psychology, 40, 273-280. Cialdini, R. B. (2001). Influence. Needham Heights, MA: Allyn & Bacon. Conlon, D. J., Carnevale, P., & Murnighan, J. K. (1994). Intravention: Third party intervention with clout. Organizational Behavior and Human Decision Processes, 57, 387-410. Coser, L. A. (1971). Masters of sociological thought: Ideas in historical and social context. New York: Harcourt Brace Jovanovich. Cummings, L. L., Huber, G.P., & Arendt, E. (1974). Effects of size and spatial arrangements on group decision-making. Academy of Management Journal, 17, 46-475. Duverger, M. (1972). Factors in a two-party and multiparty system. In Party Politics and Pressure Groups. New York: Thomas Y. Crowell. Ely, R. J. (1994). The effects of organizational demographics and social identity on relationships among professional women. Administrative Science Quarterly, 39(2), 203-238. 25 Even- and odd-sized groups even. (n.d.). Dictionary.com Unabridged (v 1.1). Retrieved January 26, 2008, from Dictionary.com website: http://dictionary.reference.com/browse/even Frank, F., & Anderson, L. R. (1971). Effects of task and group size upon group productivity and member satisfaction. Sociometry, 34(1), 135–149. Festinger, L. (1957). A theory of cognitive dissonance. Stanford, CA: Stanford University Press. Gaertner, S. L., Dovidio, J. F. (2000). Reducing intergroup bias: The common ingroup identity model. Philadelphia, PA: Psychology Press. Greenberg, J. (1966). Language universals. The Hague: Mouton. Gruenfeld, D. H., Martorana, P., & Fan, E. T. (2000). What do groups learn from their worldliest members? Direct and indirect influence in dynamic teams. Organizational Behavior and Human Decision Processes 82, 45–59. Grusky O., Bonacich P. & Webster C. (1995). The coalition structure of the four-person family. Current Research in Social Psychology, 1, 16–29. Guarnaschelli, S., McKelvey, R. D., & Palfrey, T. R. (2000). An experimental study of group decision rules. American Political Science Review, 94, 407–423. Hackman, J. R. & Vidmar, N. (1970). Effects of size and task type on group performance and member reactions. Sociometry, 33, 37-54. Hambrick, D. C. (1994). Top Management Groups: A Conceptual Integration and Reconsideration of the 'Team' Label. Research in Organizational Behavior, 16, 171-213. Hastie, R., Penrod, S. D., & Pennington, N. (1983). Inside the jury. Cambridge, MA: Harvard University Press. Hastie, R. & Kameda, T. (2005). The Robust Beauty of Majority Rules in Group Decisions, Psychological Review, 112, 494–508. Heath, C., Larrick, R. P., & Klayman, J. (1998). Cognitive repairs: How organizations compensate for the shortcomings of individual learners. Research in Organizational Behavior, 20, 1-37. Heider, F. (1958). The psychology of interpersonal relations. Wiley, New York. Hines, T. M. (1990). An odd effect: Lengthened reaction times for judgments about odd digits. Memory and Cognition, 18, 40-46. Hollingshead, A. (1998). Retrieval processes in transactive memory systems. Journal of Personality and Social Psychology, 74(3), 659-71. 26 Even- and odd-sized groups Hume, D. (1985). On the Balance of Power. In Essays, moral, political and literary. Indianapolis: Liberty Classics (Original work published in 1741). Jehn, K.A., & Mannix, E. (2001). The dynamic nature of conflict: A longitudinal study of intragroup conflict and group performance. Academy of Management Journal, 44, 238-51. Kanter, R. (1977). Men and women of the corporation. New York: Basic Books. Kreft, I. G. G., & de Leeuw, J. (1998). Introducing Multilevel Modeling. London: Sage. Lau, D. C., & Murnighan, J. K. (2005). Interactions within groups and subgroups: The dynamic effects of demographic faultlines. Academy of Management Journal, 48, 645-660. Leung, K. (1987). Some determinants of reactions to procedural models for conflict resolution: A cross-national study. Journal of Personality and Social Psychology, 53, 898-908. Levine, J. M., & Moreland, R. L. (1991) Culture and socialization in work groups. In L. B. Resnick, J. M. Levine, & S. D. Teasdale (Eds.), Perspectives on socially shared cognition (pp. 257279). Washington, D.C.: American Psychological Association. Levine, J. M., & Moreland, R. L. (1998). Small groups. In D. Gilbert, S. Fiske, & G. Lindzey (Eds.), Handbook of social psychology (4th. ed., Vol. 2, pp. 415-469). Boston: McGraw-Hill. Levine, J. M., Resnick, L. B., & Higgins, E. T. (1993). Social foundations of cognition. Annual Review of Psychology, 44, 585-612 Levine, J. M., Thompson, L. (1996). Conflict in groups. In E. T. Higgins, A. W. Kruglanski (Eds.), Social psychology: Handbook of basic principles (pp. 745-776). New York, NY: Guilford Press. Maier, N. R. (1972). Decision-making in three- versus four-person groups. Personnel Psychology, 25(3), 531-534. Mannix, E. A. (1993). Organizations as resource dilemmas: The effects of power balance on group decision making. Organizational Behavior and Human Decision Processes, 55, 1-22. Menon, T., Thompson, L., & Choi, H. (2006). Tainted knowledge versus tempting knowledge: Why people avoid knowledge from internal rivals and seek knowledge from external rivals. Management Science, 52, 1129-1144. Menon, T., & Pfeffer, J. (2003). Valuing internal versus external knowledge. Management Science, 49, 497-513. Montesquieu, C. L. (1949). The spirit of laws. (T. Nugent Trans.). New York, NY: Hafner Press. (Original work published 1748). Moreland, R. L., Argote, L., & Krishnan, R. (1996). Socially shared cognition at work: Transactive memory and group performance. In J. L. Nye & A. M. Brower (Eds.), What’s social about social cognition? Research on socially shared cognition in small groups (pp. 57-84). Thousand Oaks, CA: Sage. 27 Even- and odd-sized groups Moreland, R. L., Levine, J. M., & Wingert, M. L. (1996). Creating the ideal group: Composition effects at work. In J. Davis & E. Witte (Eds.), Understanding group behavior (Vol. 2, pp. 11-35). Mahwah, NJ: Erlbaum. Moreland, R. L., Hogg, M. A., & Hains, S. (1994). Back to the future: Social psychological research on groups. Journal of Experimental Social Psychology, 30, 527-555. Morris, M. W., Larrick, R. P., & Su, S. K. (1999). Misperceiving negotiation counterparts: When situationally-determined bargaining behaviors are attributed to personality traits. Journal of Personality and Social Psychology, 77, 52-67. Morris, M.W., Williams, Y., Kwok, L., Larrick, R., Mendoza, M.T., Bhatnagar, D., Li, J., Kondo, M., Luo, J-L, & Hu, J-C. (1998). Conflict Management Style: Accounting for Cross-National Differences, Journal of International Business Studies, 29, 729-748. Murnighan, J. K. (1978). Models of coalition behavior: Game theoretic, social psychological and political perspectives. Psychological Bulletin, 85, 1130-1153. Murnighan, J. K. & Roth, A. E. (1980). The effects of group size and communication availability on coalition bargaining in a veto game. Journal of Personality and Social Psychology, 39, 92-103. Murnighan, J. K., & Szwajkowski, E. (1979). Coalition bargaining in four games that include a veto player. Journal of Personality and Social Psychology, 37(11), 1933-1946. Newcomb, T. M. (1953). An approach to the study of communicative acts. Psychological Review, 60, 393-404. Nishiyama, Y. (2006). A study of odd- and even-number cultures. Bulletin of Science, Technology and Society, 26 (6), 479-484. odd. (n.d.). Dictionary.com Unabridged (v 1.1). Retrieved January 26, 2008, from Dictionary.com website: http://dictionary.reference.com/browse/odd O’Dell, J. W. (1968). Group size and emotional interaction. Journal of Personality and Social Psychology, 8, 75-78. O’Reilly, C. A., Williams, K. Y., & Barsade, S. (1998). Group demography and innovation: Does diversity help? In D. Gruenfeld & M. A. Neale (Eds.), Research on Managing in Groups and Teams (Vol. 1, pp. 183-207). Greenwich, CT: JAI Press. O'Leary, M.B., & Mortensen, M. (2005). Subgroups with attitude: Imbalance and isolation in geographically dispersed teams. Presented at the Academy of Management Annual Meeting, Honolulu, HI. Pelled, L. H. (1996). Demographic diversity, conflict, and work group outcomes: An intervening process theory. Organization Science, 7, 615-631. 28 Even- and odd-sized groups Peterson, R. S., Owens, P. D., Tetlock, P. E., Fan, E., & Martorana, P. (1998). Group dynamics in top management teams: Groupthink, vigilance and alternative models of organizational failure and success. Organizational Behavior and Human Decision Processes, 73, 272-305. Peterson, R. S., & Nemeth, C. J. (1996). Focus versus flexibility: Majority and minority influence can both improve performance. Personality and Social Psychology Bulletin, 22, 14-23. Pfeffer, J. (1983). Organizational demography. In L. L. Cummings & B. M Staw (Ed.), Research in organizational behavior (Vol. 2, pp. 299-357). Greenwich, C.T: JAI Press. Pfeffer, J. (1992). Managing with Power: Politics and Influence in Organizations. Boston: Harvard Business School Publishing. Phillips, K. W. (2003). The effects of categorically based expectations on minority influence: The importance of congruence. Personality and Social Psychology Bulletin, 29(1), 3-13. Phillips, K. W., Mannix, E. A., Neale, M. A., & Gruenfeld, D. H. (2004). Diverse groups and information sharing: The effects of congruent ties. Journal of Experimental Social Psychology, 40(4), 497-510. O’Reilly, C. A., Williams, K. Y., & Barsade, S. G. (1998). Group demography and innovation: Does diversity help? In M.A. Neale, E.A. Mannix, & D.H. Grunfeld (Eds.), Research on managing groups and teams (Vol. 1, pp. 183-207). Greenwich, CT: JAI Press. Rico, R., Sanchez-Manzanares, M., Gil, F., & Gibson, C. (2008). Team implicit coordination processes: A team knowledge-based approach. Academy of Management Review, 33, 163-184. Ridgeway, C. (1991). The social construction of status value: Gender and other nominal characteristics. Social Forces, 70, 367-386. Ross, L., Greene, D. & House, P. (1977). The false consensus effect: An egocentric bias in social perception and attribution processes. Journal of Experimental Social Psychology, 13, 279-301. Saks, M. J. (1977). Jury verdicts: The role of group size and social decision rule. Lexington, MA: Lexington Books. Schwartz, B. (1993). A history of the supreme court. New York: Oxford University Press. Shears, L. M. (1967). Patterns of coalition formation in two games played by male tetrads. Behavioral Science, 12, 130-137. Simmel, G. (1950). The Sociology of Georg Simmel. New York: Free Press. Slater, P. E. (1958). Contrasting correlates of group size. Sociometry, 21, 129-139. Stasser, G., Kerr, N. L., & Davis, J. H. (1980). Influence processes in decision-making groups. In P. B. Paulus (Ed.), Psychology of group influence (pp. 431-477). Hillsdale, NJ: Erlbaum. 29 Even- and odd-sized groups Sunstein, C. R., Schkade, D. A., Ellman, L. M. & Sawicki, A. (2006). Are judges political? An empirical analysis of the federal judiciary. Brookings Institution Press: Washington, D.C. Sutton , R.I. & Hargadon, A. (1996). Brainstorming groups in context: Effectiveness in a product design firm. Administrative Science Quarterly, 41, 685-718. Tajfel, H. & Turner, J. C. (1986). The social identity theory of inter-group behavior. In S. Worchel and L. W. Austin (Eds.), Psychology of Intergroup Relations. Chigago: Nelson-Hall. Thomas, E. J., & Fink, C. F. (1963). Effects of group size. Psychological Bulletin 60, 371-384. Thomas-Hunt, M. C., & K. Phillips (2003). Managing teams in the dynamic organization: The effects of revolving membership and changing task demands on expertise and status in groups. In R. S. Peterson, & E. A. Mannix (Eds), Leading and managing people in the dynamic organization (pp. 115-133). Mahwah, NJ: Lawrence Erlbaum Associates Publishers. Thorndike, E. L. & Lorge, I. (1944). A teacher's word book of 30,000 words. New York: Teachers College, Columbus University. Walt, Stephen M. (1987). The origins of alliances. Ithica, NY: Cornell University Press. Wegner, D. M. (1987). Transactive memory: A contemporary analysis of the group mind. In B. Mullen & G. R. Goethals (Eds.), Theories of group behavior (pp. 185-205). New York: SpringerVerlag. Wilk, S. L., & Moynihan, L. M. (2005). Display rule “regulators”: The relationship between supervisors and worker emotional exhaustion. Journal of Applied Psychology, 90(5), 917-927. Willis, R. H. (1962). Coalitions in the tetrad. Sociometry, 25(4), 358-376. Wittenbaum, G. M., Merry, C. J., & Stasser, G. S. (1996). Tacit coordination in anticipation of small group task completion. Journal of Experimental Social Psychology, 32, 129-152. Ziller, R. C. (1957). Group size: A determinant of the quality and stability of group decisions. Sociometry, 20, 165-173. Zimmer, K. (1964). Affixal negation in English and other languages: An investigation of restricted productivity. Word Monograph, 5. New York. 30