2008 Micro-Dynamics of Coalition Formation

8/7/2019 2008 Micro-Dynamics of Coalition Formation

1/19

University of Utah

The Micro-Dynamics of Coalition FormationAuthor(s): Daniel Diermeier, Roderick I. Swaab, Victoria Husted Medvec, Mary C. KernSource: Political Research Quarterly, Vol. 61, No. 3 (Sep., 2008), pp. 484-501Published by: Sage Publications, Inc. on behalf of the University of UtahStable URL: http://www.jstor.org/stable/20299751 .

Accessed: 03/04/2011 21:00

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless

you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you

may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at .http://www.jstor.org/action/showPublisher?publisherCode=sage. .

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed

page of such transmission.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of

content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms

of scholarship. For more information about JSTOR, please contact [email protected].

Sage Publications, Inc. and University of Utah are collaborating with JSTOR to digitize, preserve and extend

access to Political Research Quarterly.

http://www.jstor.org
http://www.jstor.org/action/showPublisher?publisherCode=sagehttp://www.jstor.org/action/showPublisher?publisherCode=utahhttp://www.jstor.org/stable/20299751?origin=JSTOR-pdfhttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/action/showPublisher?publisherCode=sagehttp://www.jstor.org/action/showPublisher?publisherCode=sagehttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/stable/20299751?origin=JSTOR-pdfhttp://www.jstor.org/action/showPublisher?publisherCode=utahhttp://www.jstor.org/action/showPublisher?publisherCode=sage


2/19

The Micro-Dynamics ofCoalition FormationDaniel DiermeierRoderick I. SwaabVictoria Husted MedvecKellogg School ofManagement, Northwestern University, Evans ton, Illinois

Mary C. KernZicklin School of Business, Baruch College, New York

We present an experimental approach to study themicro-dynamics of coalition formation in an unrestricted bargainingenvironment. Specifically, we investigate a fundamental feature of sequential coalition bargaining models: expectations about future bargaining behavior will influence current bargaining outcomes. To do so, we test the hypothesisthat coalition bargaining may lead to inefficient outcomes as agents are unable to effectively commit to preliminaryagreements during the bargaining process. We conjecture that communication plays an important role in establishingsuch commitments. We then experimentally manipulate the communication channels and show that restrictions undermine trust and lead to decreased efficiency.

Political Research QuarterlyVolume 61 Number 3September 2008 484-501? 2008 University of Utah10.1177/1065912908316981

httprZ/prq.sagepub. comhosted athttp://onlme.sagepub.com

Keywords: coalitions; dynamics; coalition formation; experiments

Duringthe past two decades, the theoretical studyof coalition governments has been transformed.

While earlier approaches largely relied on modelsadapted from cooperative game-theory (see Laverand Schofield 1990 for a detailed review of this literature), more recent approaches have used noncooperative models (Baron 1989; Diermeier and Merlo2000; Diermeier, Eraslan, and Merlo 2003; Morelli1999).l As a point of departure, consider the

Baron-Ferejohn (BF) model, the most widely usedmodel of bargaining under majority rule. In all variants of the BF model, a proposer is selected according to a commonly known rule such a randomselection proportional to seat shares. He then pro

poses a policy or an allocation of benefits to a groupof voters. According to a given voting rule, the proposal is either accepted or rejected. If the proposal isaccepted, the game ends and all actors receive payoffs as specified by the accepted proposal. Otherwise,another proposer is selected, and the process continues until a proposal is accepted or the game ends.2The BF model predicts that the party with proposalpower will propose aminimal winning coalition consisting of himself or herself and the "cheapest" set of

484

voters necessary to ensure acceptance. All other voters will receive a payoff of zero. The amount givento the coalition partners (or continuation value)equals the coalition partners' expected payoffs if theproposal is rejected and the bargaining continues.Proposals are thus always accepted in the first round.The proposing party will always choose coalitionpartners with the lowest continuation values. Thedivision of spoils will in general be highly unequal,especially if the parties are very impatient.3This description of the BF model makes it clear

why the model has been popular in the study ofcoalition formation. The "proposer" can be interpreted as the formateur. The recognition rule corresponds to the explicit or implicit selection processused by a monarch, president, or informateur. Thisselection process is usually viewed as nonpartisan.4The "voters" in the model can be interpreted asparty leaders engaging in negotiations over the formation of the next cabinet, and so forth. Note alsothat the "proposer premium" creates additionalincentives for electoral competition, as parties withhigher seat shares may be more likely to be selectedas formateur.5


3/19

The BF model was the first example of a dynamicbargaining model applied to coalition formation. Itgenerated various important and novel insights intothe specifics of coalition formation that had eludedprevious models.6 First, it emphasized the value ofproposal power. For example, it has been shown thatin a divide-the-dollar setting in which a player hasveto power but lacks proposal power, while all otherplayers have some proposal power but no veto power,the veto player receives an equilibrium payoff of zero(Diermeier and Myerson 1994). Second, the recognized proposer can increase his or her payoffsby exploiting the other players' "impatience."Interestingly, this effect is much more pronouncedthan in bilateral bargaining models (Rubinstein1982). Once agreements are subject tomajority rule,the impatience to agree to a current deal is also driven by the "concern" of being left out of a future proposal.7 That is, if a party does not agree to the currentproposal, it may not be included in future ones(Eraslan and Merlo 2002). Interestingly, in equilib

rium, this (rational) concern may lead parties to agreetoo quickly, which may lead to inefficient outcomes.Eraslan and Merlo established their result in the context of the BF model with a randomly changing pie.Intuitively, the reader may want to think of it as a"Divide-the-Euro" game with a randomly varyingexchange rate. Subsequent work, however, has shownthat this phenomenon is far more general than theEraslan-Merlo model or even the Baron-Ferejohnframework. Rather, it seems to point to a general property of bargaining under majority rule.

This can be seen in the context of an alternativeframework, so-called proto-coalition bargaining, originally proposed by B. Baron and Diermeier (2001).8Proto-coalition bargaining works as follows. A proposer

party ("formateur") is selected (e.g., proportionally toseat share). That party then selects a "proto-coalition"(Axelrod 1970), that is, a list of parties that agree to

negotiate under unanimity rule. If the proposed coalitional agreement fails to win amajority in the chamber, either a new formateur is selected and the processcontinues or a caretaker government implements anexogenously given status quo policy. The payoff associated with a particular proto-coalition is determinedby bargaining within the proto-coalition. Note thatproto-coalition bargaining is quite different from theBaron-Ferejohn protocol. In the Baron-Ferejohnapproach, a formateur not only selects a group ofparties to "talk to" but also, at the same time, a particular allocation of payoffs among the parties. Therefore,proto-coalition and coalition coincide. In proto-coalition

Diermeier et al. /Micro-Dynamics of Coalition Formation 485

bargaining, on the other hand, first, the proto-coalitionis selected, and then, in a separate step, the protocoalition agrees on a payoff distribution.Diermeier and Merlo (2000) showed how protocoalition bargaining could be used to build amodel ofcabinet stability that was able to account for all typesof coalition government (minority, minimal winning,and super-majority). Expectations of the durability ofa cabinet provided a natural interpretation of a changing pie. In other words, while parties are in govern

ment, they receive an income stream from holdingportfolios and so forth; the longer the expected duration, the bigger the total income stream. In contrast to

Eraslan and Merlo (2002), the pie does not changerandomly but reflects the common expectations aboutthe relative durability of certain government types.

Beliefs about coalition stability in turn incorporateexpectations of how subsequent coalitions will beformed if the current coalition falls. The stability andthe relative occurrence of different types of govern

ments are thus jointly determined in equilibrium. Asin the case of Eraslan and Merlo (2002), this maylead to inefficiencies.

We can summarize the two main findings ofdynamic coalition formation models as follows. First,the expectation of which coalitions (or proto-coalitions)

may form in the future if the current negotiation failscreates incentives for coalition members to extractfavorable agreements from their current bargainingcounterparts. This is true both at the stage of govern

ment formation and in the context of cabinet stability when alternative coalitions serve as the "outsideoptions" to the current agreement. These incentivescan be nested and circular. That is, an agreementwithin coalition A may depend on the expected agreement in coalition B, which may depend on theexpected agreement within coalition C, and so forth.In other words, the nature and efficiency of any currentagreement depends on the shared expectations aboutfuture agreements. Of course, these future agreements

may never come to pass if the current agreementholds?they are counterfactuals, or (in the languageof game-theory) "off the equilibrium path," but if theywere not possible, the actual agreement would bedifferent as well.

Notice that such future agreements are not onlyless favorable to current coalition partners because ofa shift in bargaining strength (e.g., proposal power

may shift to a different member of the coalition), butmost importantly, future coalitions may consist ofdifferent parties, relegating at least some of thecurrent coalition members to the much less desirable


4/19

486 Political Research Quarterly

role of opposition party. This is a defining characteristic of majoritarian bargaining: it is not necessary toreach consensus among all negotiating parties tocome to an implementable agreement. The paradig

matic case, of course, is the divide-the-dollar framework in which a simple majority can implement anyallocation. This leads to the second effect. The "fearof exclusion" not only sustains current coalitions asequilibria, itmay also lead negotiating parties to acceptinefficient outcomes out of the fear that the currentcoalition will be replaced by a new one and that they

may be left out of the final deal.The dependence of coalition outcomes on the bar

gaining protocol has led to an increased interest inempirical tests of sequential bargaining models. Onone hand, researchers have tried to establish to whatextent this modeling approach can help us understandthe effects of government formation institutions oncoalition outcomes and stability.9 On the other hand,there has been an increased interest in laboratoryexperiments to test the predictions of the sequentialbargaining approach (e.g., McKelvey 1991;Diermeier and Gailmard 2006; Frechette, Kagel, and

Morelli 2005). From amethodological point of view,the existing experimental studies in coalition bargaining follow the experimental design philosophy usedin experimental game theory. The general approach isto design an experiment that is as close as possible toa fully specified formal model to test the model'simplications. Experimentalists design ingenious

mechanisms to control for risk aversion, ensureanonymity, or implement infinitely repeated games infinite time periods. In dynamic contexts, for example,experimental participants can only use the movesspecified in the extensive form and so forth. Thisapproach is necessary if we want to test existing formal theories (i.e., high internal validity), but it ispotentially limiting if we want to understand somegeneral

features of bargaining that either have notbeen fully captured by formal models or are characteristic for a whole research approach (i.e., low external validity).

Our goal in this article is not to test a specific bargaining protocol (e.g., the BF model) but to empirically assess general implications of the sequentialbargaining approach no matter how the models arespecified in detail. Our main concern is to assess therelationship between a "fear of being left out" andoutcome inefficiency. Since this is perhaps the mostsurprising implication of the sequential bargainingapproach, an affirmative finding would present considerable support for the general approach.

In contrast to the recent experimental tradition, weintentionally do not specify an exact sequence of

moves, nor do we tightly restrict the ability of participants to communicate. Rather, participants are freeto apply various strategies to solve the coalition for

mation problem in an unrestricted setting. Most importantly, we allow participants to talk to each otherwithout any restrictions. In some setups, they are ableto talk privately, ask other parties to leave the room,and so forth. That is, our purpose is to (re-) create inthe lab as rich a negotiation environment as possible.

Being able to identify some of the main features ofsequential bargaining in such unrestricted settingsallows us to add external validity to the sequential bargaining approach. Our approach, while motivated bya formal model, is more in line with most socialpsychology experiments in which experimentalistsare trying to test the validity of theoretical insightsrather unrestricted settings.

The Fear of Being Left OutOur experiments are based on the following nego

tiation problem among three parties (A, Z?,and C; seeTable l).10 Depending on the agreement, differentcoalitions receive fixed amounts of money that can befreely distributed among the parties.An agreement consists in an allocation of the respective amount among the parties involved in the deal.For example, any agreement between A and B alonewould specify both A's and B's share of the specified$118,000. If no agreement can be reached, each partyreceives a payoff of $0.nIn this example, an efficient outcome involves theparties' deciding to leave no resources unused, whichimplies that all parties need to be included in the deal.Of course, each party has different preferences abouthow the total pie should be split. What makes thisproblem difficult is that any two parties (e.g., A and B)can form a moderately profitable agreement withoutincluding the third party (Q. So, one possible intuitionof how the negotiations may proceed is as follows.Parties A and B (or some other "proto-coalition") mayform a preliminary agreement on how to split the piealready available to an A-B coalition (here, $118,000)between themselves and then only negotiate over theremaining amount ($3,000) with C. This scenario,however, is highly unfavorable to C, who will thus tryto break up any proto-coalition between A and B toavoid being left with a pittance. Given the negotiation'scollective incentive structure, C can make an attractive


5/19

Table 1Payoff Matrix

Total Payoff (ToBePossible Distributed amongAgreements Parties inAgreement)

A alone $0B alone $0C alone $0 A and B $118,000A and C $84,000B and C $50,000A, B, and C $121,000

offer to either A or B, regardless of how A and B havedivided the $118,000 between themselves. Forexample, ifA and B decide to split the money equallybetween themselves (i.e., $59,000 for each A and E),C can offer A at least $60,000 (while keeping $24,000for herself). The same holds in any agreement in whichB gets more of the $118,000, or, analogously, in thecase inwhich A receivesmore of the $ 118,000 but lessthan$78,000. IfA receives $78,000 ormore (and thus,B receives $40,000 or less), C can offerB $42,000 andkeep $8,000 for herself. If party C is successful intempting either A or S to form a new proto-coalition,the whole process begins anew.12

The key insight is that an advantageous strategyfor a player is to form a bond with one other player andthen jointly take advantage of the weak bargainingposition of the third. In the context of coalition bargaining, such a bond corresponds to a proto-coalition.This approach is complicated by the fact that the thirdplayer can always make a sufficiently attractive offerto induce one of the members of the proto-coalitionto defect from the preliminary agreement. Indeed,

maintaining the current proto-coalition while negotiating with the third party over the remaining surplusconstitutes the core difficulty of this bargaining problem. Inefficient outcomes may result if players are"too concerned" with maintaining the current protocoalition to successfully come to an agreement withthe third party. The result is an inefficient outcome:

money will be left on the table. Moreover, the mereconcern that the other party may defect may lead aparty to abandon the current proto-coalition.13The goals for negotiating parties are thus to (1) beincluded in the proto-coalition and (2) ensure that theproto-coalition is stable. Conversely, in contrast tobilateral negotiations, the main risk is not an impassebut being "left out" of the deal (i.e., either receivingnothing or only a pittance). According to our definition,


inefficiency may correspond to an impasse (i.e., noagreement) or any two-party agreement.

The micro-dynamics of coalition bargaining aresummarized in Figure 1.In the discussion of this phenomenon,we have

introduced psychological terms such as fear and concern. While they are foreign to the rational actor modelused in game-theory, we expect them to play an important role in real bargaining situations. The strategicrequirement to build stable proto-coalitions corresponds to the need to form a trusted bond between the

members of the proto-coalition, as formally bindingcommitment devices or contracts are not available. Inother words, the less trusted a proto-coalition, themoreeffort is necessary for its members to stabilize it andthe lesswilling they are to include the thirdplayer inthe discussions, as they always may fear that the thirdplayer may take advantage of the invitation to negotiate by breaking up the existing proto-coalition.For stable proto-coalitions to occur, parties thereforemust be able to establish a sense of trustwith their protocoalition partners. Trust can be established in manydifferent ways. For example, members of the protocoalition may reject offers from the third party eventhoughthe shareofferedby the thirdparty ishigher thantheir current share in the proto-coalition agreement.

Alternatively, members of the proto-coalition may agreeto communicate only jointly with the third party.These and other techniques to establish mutualtrust depend on preconditions. For instance, players

may rely on nonverbal communication to signal anagreement (i.e., they may seek eye contact beforespeaking to the third party or may move together onthe same side of the table). Thus, the degree to whichnonverbal factors can be used (e.g., in a face-to-faceversus computer-mediated negotiation) may influencethe stability of proto-coalitions and whether or notsomeone is excluded from the final outcome. Also, thesignaling effect of rejecting

an attractive offer by theoutside party presupposes that outside offers (andtheir rejection!) can be observed by all of the parties;they must be public. Similarly, agents may be concerned that their proto-coalition partner can secretlycommunicate with the third party, perhaps negotiatinga more attractive deal on the side. Note that even thepossibility of secret communication may undermine aparty's trust in the current proto-coalition.This suggests that the communication channels parties have available strongly influence the trust betweenmembers of proto-coalitions and whether they can beexpanded to fully efficient three-player coalitions. Ourempirical strategy is as follows: rather than trying to


6/19


Figure 1Proto-Coalition BargainingA: By definition, players begin negotiating in a

"grand coalition" setting. Theythen can decide

on whether to form a proto-coalition or theycan negotiate a final outcome (this may includereaching an impasse) within the entire group.B: If two players decided to form a proto-coalition,they negotiate with each other until they reachan agreement or reach an impasse.

C: The agreement may only contain two playersand no further negotiations take place, inwhichcase this will be the final outcome.

D: Alternatively, if the two players in Phase IIdecide to include the third player in the negotiation, they move to Phase III.E: Finally, if the two players in Phase II disagreein their negotiation, they can dissolve the protocoalition and return to Phase I.

F: If the third party breaks up the existing protocoalition, they return to Phase I.

G: If the now expanded proto-coalition reaches anagreement, this will constitute the final deal.

measure or directly manipulate existing trust betweenproto-coalition partners, a rather difficult challengeunder any circumstances, we manipulate the communication channels that are used during the coalition negotiations. We summarize our argument in Figure 2.

Notice that our approach depends on two crucialtheoretical steps, here denoted (A) and (B). The relationship between commitment and trust followsdirectly from our discussion of the dynamic-bargainingliterature. It is the main object of investigation.However, since we cannot measure trust and commitment within the (several) proto-coalitions directly, weuse variations in the communication channel instead.The validity of this approach depends on a separatehypothesis relating communication channels withtrust and commitment (A). This connection has beenextensively investigated in the communications literature to which we turn next.

The Impact of CommunicationChannelsCommunication channels vary in the amount of

cues they may carry, which in turn may affect the

Figure 2Communication-Trust-Efficiency

Communication ... ;Trust and i ._. Outcome EfficiencyChannel ' ' 11Commitment i \ / t two versus threeverbal / nonverbal verbal / nonverbal | players?public / secret j*public / secret j |? impasse

Note: Solid lines indicate attributes measured in our study.

success of communication processes (Daft andLengel 1986). This "channel richness approach" tomedia effects has been central to many theories,

including Social Presence Theory (Short, Williams,and Christie 1976), Information Richness Theory(Daft and Lengel 1986), the Cuelessness Model(Rutter 1987), and the Reduced Social Cues Approach(Kiesler, Siegel, and McGuire 1984). Although thereare qualitative differences in the primary foci of thesetheories, all argue?in one way or another?that a

medium's capacity to convey several cues at the sametime determines whether or not people are able towork effectively with one another. That is, the morecues the medium conveys, the more likely

the interaction will be successful.


7/19

Some of the negotiation research supports the ideathat negotiation outcomes may be a function of theamount of cues a medium conveys. Compared to textbased communication, for example, the presence ofaudio cues may contribute to fewer impasses(Mennecke, Valacich, and Wheeler 2000; Purdy and

Nye 2000; Suh 1999;Valley, Moag, and Bazerman1998). Also, the concurrent presence of audio and visualcues during face-to-face negotiations has contributed tomore efficient outcomes than text-based communication(e.g., Croson 1999; Naquin and Paulson 2003;

Rangaswamy and Shell 1997) or audio-only communication (e.g., King and Glidewell 1980;Morley andStephenson 1969, 1970; Short 1974). The underlyingreason suggested in this body of research is that thepresence of more cues reduces ambiguity and positivelyrelates to one's ability to exert influence, making theinteraction process more "human" and inherently moresocial. Although this research suggests that similareffects are to be expected in coalition formation, thisremains an open question as prior research exclusivelyfocused on bilateral interactions. We capture thisapproach by comparing face-to-face communicationwith various forms of computer-mediated communication, which, in our setting, is entirely text based.

The channel richness approach has not been without challenges. One argument is that capacityapproaches ignore the importance of other contextualfactors underlying media use (Walther 1996). Contraryto capacity approaches, these "contextual perspectives"argue that the effects of communication channels arelargely shaped by people, experience, and the interaction between them (see, for example, Carlson andZmud 1999; Fulk 1993). According to this perspective, group processes and outcomes are determinedby the setting in which communication takes placerather than by technological factors alone.Recent findings from the negotiation literature support this contextual perspective. For example, a recentlyconducted meta-analysis on the impact of communication channels on negotiations shows that the contribution of multiple cues to negotiation outcomes isinfluenced by a negotiator's relationship with the counterpart or the experience one has with the technology(Swaab, Medvec, and Diermeier n.d.). An additionalcontextual factor often ignored in this literature is theextent towhich communication settings allow people tobe aware of the communication content. Little is knownabout the effect of such public awareness on groupprocesses in general and multiparty negotiations inparticular. This is somewhat surprising given theconsequences that awareness of the conveyed message


may have for decision-making groups who maystrongly depend on having an idea of who is communicating with whom. For example, during multiparty faceto-face interaction, negotiators can monitor thecommunication occurring between various parties.Either the communication is public and all of the partiespresent hear the discussion, or if private communicationtakes place during the meeting, the parties generallyknow that this is occurring (they see two parties whispering to each other, hear them speak in a foreign language or unknown code, or know that two parties leftthe room together).14In summary, we expect that decision makingprocesses and outcome efficiency can be affected bythe communication channel (i.e., face-to-face or computer mediated) as well as the communication setting(the extent to which the content of the communication is accessible to all involved). That is, we willinvestigate the following theoretical claims:

Theory 1: Compared to computer-mediated communication, face-to-face communication will

lead tomore efficient outcomes in coalition formation problems.

Theory 2: Compared to private communication settings, public communication settings will lead tomore efficient outcomes in coalition formationproblems.

Experiments: Design and ResultsWe examine these predictions in three experimental studies. The first study examines the differences

between face-to-face and computer-mediated decisionmaking. The second study is designed to replicate theeffects of the first study and to examine the impact ofpublic versus private communication settings on outcome efficiency and on how participants frame theirdiscussions (e.g., in terms of group versus subgroupinterests). The third study isolates the effects of thissecond factor and disentangles the differencesbetween private and secret communication settings inoutcome efficiency and communication framing. Wewill use the coalition formation scenario described inthe introduction as well as a common research designthroughout the three experimental studies.

Study 1Study 1 is designed to test the impact of synchronous text-based computer-mediated versus face-to-face

communication on outcome efficiency in amultiparty


8/19


negotiation. Based on previous computer-mediatedcommunication research that focused on group decision

making (Bakes et al. 2002) and dyadic negotiations(e.g. Arunachalam and Dilla 1992; Croson 1999;

Morris et al. 2002; Naquin and Paulson 2003;Rangaswamy and Shell 1997; Suh 1999),we hypothesize that the multitude of cues present in face-to-facecommunication makes it easier to sustain trust in theother member of the proto-coalition. This will makeitmore likely that the members of the proto-coalition

will risk expanding the coalition compared to themembers in the computer-mediated communication(CMC) scenario. The following hypothesis capturesour first claim (Theory 1).

Hypothesis 1:Groups using computer-mediated communication are less likely to reach efficient outcomes than groups communicating face-to-face.

MethodParticipants and design. Eighty-seven students

participated in this study as part of a course on executivebusiness administration negotiations at aMidwesternUnited States business school. The participants wererandomly divided into twenty-nine three-persongroups. The experiment was a between-subjects designwith communication medium (face-to-face versusCMC) as the independent variable.

Task. The students participated in a three-partydecision-making case titled "Social Services." In thisnegotiation simulation, participants were randomlyassigned to the role of a representative for one of threedifferent social services agencies (Allied, Benevolent,and Caring). The agency representatives were offeredthe opportunity to receive funds from a governmentagency, provided that at least two of the three firmsformed a consortium. Each representative was giventhe goal of obtaining as much funding as possible forhis or her organization.The students were told they needed to agree on theparties to the consortium and the distribution of theresulting funds. The amount of resources available forany grant varied based on the agencies involved (seeTable 1 for the payoff matrix). All of the participantsreceived the same case materials and saw the samepayoff matrix. This fact was publicly announced atthe beginning of the exercise. As is customary inpsychology experiments, participants were confrontedwith a hypothetical-choice situation (e.g., Kahnemanand Tversky 1979; Arkes, Herren, and Isen 1988).

Following this approach, participants were not paidand did not receive special course credit contingent onnegotiation success. There was no future or follow-upinteraction after the participants' completion of thenegotiation.

Procedure. Ten of the groups were told they wouldcomplete this exercise face-to-face, and the remaining nineteen groups were told they would completethe exercise in an online chat room (the computermediated communication (CMC) condition). Theonline chat room allowed for synchronous, publiccommunication among all three parties. Both mediaalso allowed for private communication between anytwo of the three parties. In the face-to-face treatment,parties could ask one person to leave the room to conduct a private conversation. In the case of CMC,students could also send each other instant messageswithout the knowledge of the third party. The onlineexercise was conducted using the software packageAye Ware.15

The participants were allowed ten minutes to readthe exercise and forty-five minutes to come to anagreement.16 The face-to-face negotiations werevideotaped, and we obtained the transcripts from thecomputer-mediated negotiations (including instant

messaging transcripts). After completing the negotiation, participants returned to the classroom and weredebriefed as part of the course lecture (Loyd, Kern,andThompson 2005).

Communication-channel manipulation. Twoforms of communication media were used: face-toface communication and CMC. Participants in theface-to-face condition were assigned to breakoutrooms with their respective group members. Theywere told not to communicate with any other groups.Participants in the CMC condition were assigned toindividual computers in small breakout rooms, inwhich they logged on to Aye Ware. They were notallowed to verbally communicate with anyone whileengaged in the online negotiation, and there were noother members of their negotiating group in theirbreakout room.

Efficiency. We measured efficiency as the maximaluse of the total resource pool ($121,000). To fullydistribute the available pool, all three group memberswould need to agree to the deal. Efficiency wasoperationalized as a dichotomous variable?the agree

ment was coded as being either efficient or inefficient.17 This is without loss of generality. There were


9/19

Figure 3Efficiency ofMultiparty Negotiation Outcomes

100%

Face-to-face CMCNote: CMC = computer-mediated communication.

no groups that agreed on a three-player coalition butthen decided to only partially allocate the additionalsurplus.

ResultsAll analyses were conducted at the group level.

Hypothesis 1 predicted that groups would be morelikely

to reach efficient outcomes when communicating face-to-face than in chat rooms. We ran chisquare analyses comparing the efficiency of CMC toface-to-face negotiations to test this hypothesis.Supporting Hypothesis 1, groups negotiating face-toface were significantly more efficient than groupsnegotiating via CMC (Mann-Whitney U = 38.50;Z = -3.23; p = .001). In the face-to-face condition,70 percent of outcomes were efficient, as compared toonly 11 percent in the CMC condition.These results are not caused by a high frequency ofnegotiation impasses. Indeed, out of twenty-nine negotiations, only one resulted in an impasse (in the CMCcondition). Rather, in the CMC condition, participantswere almost always able to form two-party coalitionsbut failed to reach three-party agreements.18

DiscussionAs we predicted, groups reached more efficient outcomes when interacting face-to-face than in a com

puter- mediated chat room. The increased efficiencyin face-to-face negotiations may be caused by anumber of different factors. For example, when negotiating face-to-face, parties can witness and managethe formation of coalitions; they can see when a


coalition partner might defect, and they can anticipatepotentially getting shut out of a final agreement. Inaddition, itmay be more difficult tomonitor the otherparties' interactions in the CMC negotiation setting.This may make itmore challenging

to reach an efficientoutcome, as members of an existing proto-coalition

may be reluctant to communicate with the remainingparty out of a concern that this interaction may disrupt the existing agreement. Finally, is the impact offace-to-face communication the result of channelrichness, communication context, or both?

Let us first consider the potential impact of communication. To distinguish between channel richnessand communication context, we need to focus on someof the more subtle differences between computer

mediated and face-to-face interaction. One importantdifference between the face-to-face and CMC negotiations in Study 1 was that face-to-face communication not only allowed nonverbal communication butalso precluded a particular form of private communication. The lack of nonverbal cues relates to thechannel-richness hypothesis. To see the effect ofcommunication context, recall that in the CMC negotiations, parties could communicate via instant messaging. While instant messaging is an important featureof real-life online communication, it does add another,potentially important dimension to the problem. WhenParty A sends an instant message to Party B (but notto Party Q, not only is the content of Party A's message to Party B unknown to C, but Party C does noteven know that Parties A and B have communicatedprivately. That is, both the content of the communication and the fact that communication took place isprivate information held only by the communicatingparties (here, A and B). We refer to this form of communication as "secret." Hiding the fact that privatecommunication is taking place, however, is impossible in face-to-face communications. The only way forA and B to communicate privately during a face-to-facemeeting is for both of them to leave the room, to askC to leave the room, or towhisper to each other in frontof C.19 In other words, while the content of privatecommunication is hidden to C, the fact that A and Bare communicating is common knowledge among allthree parties. For simplicity, we will call this form ofcommunication "private."The conceptual difference between secret andprivate communication allows us to study some fairlysubtle aspects of the communication-context approach.Conceptually, the difference between public, private,and secret communication is what aspect of communication is common knowledge among the participants:


10/19


Table 2Communication Context and Common Knowledge

Fact That Private Communication Took Place IsCommon KnowledgeYes No

Content of private communicationis common knowledge

Yes Public-communication condition Not possible(only common chat room)

No Private-communication condition Secret-communication condition(private chat room available) (instantmessaging available)

the fact that the private communication took place orthe content of the communication setting. This issummarized in Table 2.This insight is important for the interpretation of our

findings. Research in game-theory (Aumann 1976;Rubinstein 1982;Morris and Shin 1999) has demonstrated that communication environments that preventthe establishment of common knowledge may makecoordinated, risky action impossible.20 In other words,secret communication may further reduce efficiencycompared to private communication, since with secretcommunication, proto-coalition members can neverbe sure that they are not (secretly) being doublecrossed. This is likely to further undermine trust in theproto-coalition partner and make the two parties in aproto-coalition even more reluctant to continue tonegotiate with the third party once they have achievedan acceptable agreement between themselves. Thus,we would expect multiparty CMC negotiations withsecret communication available (such as instant messaging) to result in less efficient outcomes than multipartyCMC negotiations with only private communicationavailable (private chat rooms). In contrast, we predictthat efficiency will be highest when only public communication is allowed (public chat rooms). Note,

however, that according to the channel-richnesshypothesis, even when controlling for private communication, multiparty face-to-face negotiations wouldstill result in lower rates of outcome efficiency thanany multiparty CMC negotiations.To analyze these various hypotheses in detail, weconducted two studies. Study 2 will allow us to directlytest the channel-richness and the communicationcontext theories separately and also to assess theirpossible interaction. Study 2 will also allow us tooperationalize the concept of proto-coalitions moredirectly. Study 3 will then compare private chat roomswith instant messaging to study the effects of lack ofcommon knowledge (i.e., private versus secret communication) inmore detail.

Study 2Study 2 isdesigned to separately test (1) the effectsof the channel-richness and communication-context

hypothesis and (2) investigate the proto-coalitionbargaining structure in more detail. Both goals areaccomplished by restricting communication to eithera public room (a chat room in the CMC case, ameeting room in the face-to-face condition) or allowingfor private rooms (a private chat room, or in the faceto-face condition, the ability to ask the third party toleave the meeting room and wait outside). The 2x2design will also allow us to study potential interactionbetween the two hypotheses.

We expect that the availability of private chatrooms will decrease efficiency. The intuition is thatproto-coalitions that lack sufficient trust in eachother can now essentially shut out the third party bystaying in the private chat room. Although this is truein both face-to-face and computer-mediated communication, we expect this effect to be weaker in faceto-face settings because itmay be harder to excludesomeone from a face-to-face meeting than from adiscussion cloaked by the screen of the Internet. Thisleads to the following hypotheses. Note that

Hypothesis 2 is analogous to Hypothesis 1, butin a setting in which instant messaging was notpermitted.

Hypothesis 2: Computer-mediated communicationwill lead to less efficient outcomes compared toface-to-face communication.

Hypothesis 2a: Computer-mediated communicationwill lead to less efficient outcomes compared toface-to-face communication if only public communication is permitted.Hypothesis 2b: Computer-mediated communicationwill lead to less efficient outcomes compared to

face-to-face communication if private communication is permitted.


11/19


Hypothesis 3: Private communication will leadto less efficient outcomes compared to publiccommunication.

Hypothesis 3a: Private communication will lead toless efficient outcomes compared to public communication in computer-mediated negotiations.Hypothesis 3b: Private communication will lead to

less efficient outcomes compared to publiccommunication in face-to-face negotiations.

The availability of private (chat) rooms also allows ustomore directly assess the proto-coalition bargaininghypothesis. In Study 1, it was difficult to assesswhether negotiators did indeed follow the patternsummarized in Figure 1 (i.e., moving from a grandcoalition setting to a proto-coalition setting and thenpossibly expanding the coalition again). In Study 2,we can operationalize private chat rooms as a proxyfor proto-coalition. That is, since proto-coalitions arenothing else but agreements to talk to each other (butnot to parties outside of the proto-coalition!), it isnatural to assume that newly formed proto-coalitionswould take advantage of a private chat room if such aroom were indeed available. While it is certainly possible to form proto-coalitions in a purely public setting,most drastically by ignoring the thirdparty, there islittle to be gained from staying in a public setting if aprivate setting is available.

Hypothesis 4: Two-person proto-coalitions are morelikely to occur when the communication settingallows for private communication.

Hypothesis 4a: In computer-mediated communication, two-person proto-coalitions are more likelyto occur when the communication setting allowsfor private communication.

Hypothesis 4b: In face-to-face communication,two-person proto-coalitions are more likely tooccur when the communication

settingallows

for private communication.

MethodParticipants and design. One hundred fifty-sixstudents composing fifty-two groups participated inthis study as part of a negotiations course exercise inaMasters of Business Administration program. The

experiment was a 2 x 2, between-subjects designcrossing communication medium (face-to-face versus CMC) with communication setting (private com

munication versus public communication).

Procedure. The task was identical to that used inStudy 1. Individualswere informed that theywould beparticipating in a negotiation in which each of themrepresented a social services agency. Seventeen of thegroups were told they would complete this exerciseface-to-face, and the remaining thirty-five groups weretold they would complete the exercise in an online chatroom. To control for whether the opportunity for secretcommunication played a role in the results of Study 1,we used a feature of the AyeWare technology thatallows the experimenter to set up various "private chatrooms" (i.e., chat rooms that only a subset of the negotiators can access). Intuitively, any two parties couldleave the "public chat room" and continue their negotiation in a "private chat room." However, if any partyleft a chat room, the message "[Screen Name] has leftthe chat" appeared on each party's computer screen.

AyeWare also permits us to disable instant messaging.These features allowed us to replicate a key feature offace-to-face negotiations in the online environment: ifany two parties left the public chat room, the contentof their communication may have been private, but thefact that they had left was common knowledge (exactlyas in the face-to-face condition).

Private communication was enabled for half of theparticipants in the face-to-face and CMC conditions.In the face-to-face condition, participants with theability to communicate privately were allowed to askthe third party to leave the room, while in the CMCcondition, participants with the ability to communicate privately could go to a private chat room (i.e., achat room for A and B only, a chat room for B andC only, or a chat room for A and C only). However,parties were not able to send instant messages. Theother half of the participants in each condition wereonly able to communicate publicly.The negotiation task was presented to the participants and included background information and thepayoff matrix. Participants were advised that all information was commonly shared. The participants weretold that they would receive no funding if they did notform a consortium with at least one other agency.

The participants were allowed ten minutes to readthe exercise and forty-five minutes to negotiate. Theface-to-face negotiations were videotaped, and weobtained the transcripts from the CMC negotiations.After completing the negotiation, participants returnedto the classroom to be debriefed.

Communication-channel manipulation. As in Study 1,two forms of communication media were used:


12/19


face-to-face communication and computer-mediatedcommunication.

Communication-setting manipulation. Half of thegroups in the face-to-face and CMC conditions wereallowed to communicate privately.

In the face-to-face communication context, theywere told:If two of the three organizations wish to speak privately, the third organization may be asked to leavethe room.

In the CMC conditions, the groups were told:You have two opportunities to communicate with theother

representatives?youcan chat in the common

public chat room, or you can chat in a private chatroom with either one of the two other agencies.To access the private chat room (if two of the threerepresentatives wish to speak privately), go back toyour "My AyeWare" page by hitting either the "Back"button or the "My AyeWare" button. Click on thespecific subgroup to talk to either of the other tworepresentatives. (For example, if you are Allied andwould like to talk to Benevolent, go to the subgrouplabeled "AB." If you are Caring and would like totalk to Benevolent, go the subgroup labeled "BC")The third party will not have access to the information exchanged in this private chat room.Those groups in the face-to-face condition that

were not allowed private communication were toldthat all three parties had to remain in the room at alltimes. In the CMC condition, those parties that werenot allowed private communication were not givenaccess to any private chat rooms, so all of their communication took place in the public chat room.

Efficiency. Efficiency was measured exactly in thesame way as in Study 1.Proto-coalition activity. To assess proto-coalition

activity, we coded videotapes and transcripts forwhether or not parties engaged in private conversations (1) or not (0).Results

As in Study 1, we found a very large difference inoutcome efficiency between face-to-face and CMCconditions, supporting Hypothesis 3 (Mann-WhitneyU = 172.00; Z

=-2.83; p

=.005). That is, acrossthe two face-to-face conditions, 76 percent of the

Figure 4Efficiency ofMultiparty Negotiation Outcomes

100%or| g 80%

F2F-private F2F-public CMC-private CMC-public

Note: CMC = computer-mediated communication; F2F = face-toface.outcomes were efficient as compared to 34 percentacross the CMC conditions. Since instant messagingwas not available in Study 2, we can conclude that efficiency decreases if private communication is possible,even if the fact that such communication took place iscommon knowledge. No support was found for

Hypothesis 2a. That is, the difference in efficiencybetween the public face-to-face condition (71 percent)and the public CMC condition (50 percent) was notsignificant (Mann-Whitney U = 49.50; Z = -.95; ns).

Hypothesis 2b, however, was supported. Efficiencywas higher in the private face-to-face condition (80percent) than in theprivateCMC condition (18percent;

Mann-Whitney U = 32.00; Z = -3.13; p = .002).Although there was a trend that private communication (49 percent) led to lower outcome efficiency than

public communication (61 percent), this difference wasnot significant (p = .14), leading us to reject Hypothesis3. On closer inspection, however, in the case of computer-mediated communication and consistent withHypothesis 3a, the availability of private chat rooms diddecrease efficiency in the CMC condition. That is, CMCgroups in which private communication was allowedreached efficient outcomes only 18 percent of the time,

whereas groups with only public communicationreached efficient outcomes in 50 percent of the cases(Mann-Whitney U = 103.50; Z = -1.99; p = .047). This

finding is in contrast with the face-to-face condition, inwhich allowing for private communication did not havea significant effect (p = .69), leading us to rejectHypothesis 3b.21 Figure 4 illustrates our findings.

Hypothesis 4 was also supported. Whereas protocoalitions did not occur (0 percent) in the publiccommunication conditions, they occurred in 74 percent


13/19

of the groups in the private-communication conditions (Mann-Whitney U = 87.50; Z = -5.43; p < .001).We also found support for Hypothesis 4a. That is,two-person proto-coalitions were significantly morelikely in the CMC private condition (100 percent)than in theCMC public condition (0 percent;Mann

Whitney U = .00; Z = -5.83; p < .001). Hypothesis4b was not supported. Although two-person protocoalitions were descriptively more likely in the faceto-face private condition (30 percent) than in theface-to-face public condition (0 percent), this difference was not statistically significant (Mann-WhitneyU = 24.50; Z = -1.55; p = .12).

DiscussionAs in Study 1, we found that face-to-face negoti

ations resulted in more efficient outcomes thancomputer-mediated negotiations. This provides additional evidence for the channel-richness hypothesis ina more controlled setting. Study 2 also revealed thatprivate communication further decreased in CMCnegotiations but not in face-to-face interactions.Importantly, Study 2 established that this differenceoccurs not only because of the suspicion of ongoingsecret negotiations in CMC induced by instant messaging but even when the fact that private communication is taking place becomes common knowledge.In the face-to-face setting, however, we did not identify a significant decrease in efficiency caused by theavailability of private communication. We also foundless willingness to exclude the third party from thenegotiations than expected. There are various possibleexplanations for why such an effect could not beobserved. One possible explanation is that when communication is face-to-face, participants may becomemore reluctant to exclude others from the communication process. In a face-to-face setting, the physical presence of the left-out party may put pressure on the

members of the proto-coalition to continue negotiatingwith him or her. In this setting, the left-out party iseither sitting across the table or waiting outside theroom, and the party's presence is salient and createssocial pressure. However, in the online setting, the leftout party does not have the ability to "remind" theproto-coalition that he or she is present. Thus, the effectof physical presence suggests that in face-to face interactions, the proto-coalition members will be embarrassed to "shut the door" on the third partypermanently, as this may be considered socially unacceptable behavior. This sense of reluctance or restraint,


however, may be much less constraining in onlinesettings.22 While the explanation certainly sounds plausible, previous research also suggests that computer

mediated interactions can be highly social and subjectto pressures similar to those we experience in face-toface communication (e.g., Swaab, Medvec, andDiermeier n.d.).We can test this explanation in the following

hypothesis using the data from Study 2.Hypothesis 5: Two-person proto-coalitions aremore likely when communication is computer

mediated communication compared to faceto-face communication.

Hypothesis 5a: Computer-mediated communicationwill lead to more two-person proto-coalitionscompared to face-to-face communication ifonly public communication is permitted.

Hypothesis 5b: Computer-mediated communicationwill lead to more two-person proto-coalitionscompared to face-to-face communication ifprivate communication is permitted.

Hypothesis 5 was indeed supported. Two-personproto-coalitions were more likely in computermediated communication (49 percent) than in faceto-face communication (19 percent; Mann-WhitneyU = 205.50; Z = -2.13, p = .03). Hypothesis 5a wasnot supported. That is, two-person proto-coalitions didnot occur in theCMC public condition (0 percent) orin the face-to-face public condition (0 percent), p = 1.However, we did find support forHypothesis 5b suchthat two-person proto-coalitions were more likely inthe CMC private condition (100 percent) than in theface-to-face private condition (33 percent; Mann

Whitney U = 25.50; Z = -3.93; p < .001).To further assess the validity of this approach, wereturned to Study 1 to test the analogous hypothesisin that setting. Recall that in Study 1, both private andsecret communication was permitted. We then havethe following hypothesis:Hypothesis 6: Two-person proto-coalitions are more

likely to occur when communication is computer mediated than when it is face-to-face.

Hypothesis 6 was also supported. Proto-coalitionsoccurred more often when communication was computer mediated (100 percent) than when communication was face-to-face (30 percent; Mann-WhitneyU = 28.50; Z = -4.11; p < .001).


14/19


Overall, this suggests that public-communicationconditions (whether in the CMC or the face-to-facesetting) strongly discourage the development of twoperson coalitions. If private communication is possible(e.g., by the availability of

aprivate room), negotiatorswill take advantage of it; however, this tendency

appears to be mitigated by the social pressure not toexclude a person in a face-to-face setting. This suggeststhat psychological mechanisms (here, the aversion ofsocial exclusion in the physical presence of a negotiator) may mitigate strategic incentives.

Study 3To understand the role that the communication set

ting plays in online interactions, Study 3 attempts toisolate the effects of private and secret communication.Study 1 showed the independent effects of secretcommunication in the online setting, while Study 2showed the effects of private communication; inStudy 3, we will directly compare these effects.Recall that to move toward a three-party agree

ment, members of existing proto-coalitions need tobe willing to interactwith the third party. But thatwillingness will be undermined by a concern that thethird party may want to sabotage the existing agreement. We hypothesize that the extent of this concerndepends on the ability of proto-coalition members tomonitor ongoing communication. When secret communication is possible, parties are completely unawareof interaction, and therefore unable tomonitor com

munication that is taking place between other parties.Importantly, the mere suspicion that secret communication is taking place may completely undermine the

willingness to continue negotiating with a third partyonce an initial proto-coalition agreement has beenformed. In private communication, on the other hand,

while the content of the communication remainsunknown, the fact that parties are communicating iscommon knowledge and is therefore easier tomonitorand potentially counteract.

Hypothesis 7: Outcome efficiency will be lower ingroups negotiating via CMC with the ability tocommunicate secretly than in CMC groups withthe ability to communicate privately.

MethodParticipants and design. Seventy-five students

composing twenty-five groups participated in thisstudy as part of a negotiations course exercise in a

Masters of Business Administration program. Theexperiment was conducted entirely online and manipulated the type of private communication allowed(secret versus private).

Procedure. The task was identical to that used inStudies 1 and 2. Participants were informed that theywould be involved in a negotiation in which each ofthem represented a social services agency. All twenty

five groups completed the negotiation online usingAyeWare. Communication was enabled in one of twomethods. Some of the groups were offered the opportunity to use secret communication (via instant messaging), by which they could send inconspicuous

messages to the other parties while also negotiating inthe public chat room. Other groups were provided withprivate chat rooms, where they could meet with oneother party. However, to access a private chat room, theindividual had to leave the public chat room, and thisdeparture would be commonly known by all parties.

Communication-setting manipulation. Elevengroups had access to instant messaging, and fourteengroups had access to private chat rooms.The eleven groups with access to instant messagingwere told,

You have two opportunities to communicate with theother representatives?you can chat in the commonpublic chat room, or you can send an inconspicuous

message to another party while remaining in the common chat room using instant messages. To access theinstant messaging feature (to send an inconspicuousprivate message to another party), click on the "Instant

Message" button on the lower left-hand corner of thescreen and select the party you wish to send a message to?the third party will not have access to theinformation exchanged, nor will they know that aninstant message was sent or that an instant messagewas received by any other party.

The fourteen groups with access to private chatrooms were given the exact same manipulation as inStudy 2. Outcome efficiency was measured the sameas in Study 1 and 2.Results

Hypothesis 7 predicted that groups with access tosecret communication (i.e., instant messaging) would beless likely to reach efficient outcomes compared to groupswho only had the ability to communicate privately, andthis prediction was supported (Mann-Whitney


15/19

U = 49.50; Z = -2.17; p = .03). In the experiment, nota single group was able to reach an efficient outcome,compared to 36 percent of the groups allowed privatecommunication.

DiscussionThe ability to communicate secretly had by far the

largest negative effect on outcome efficiency inmultiparty negotiation. Secret communication led to significantly fewer efficient outcomes compared to privatecommunication. The results from the first study areconsistent with this finding: only two groups in theCMC condition inStudy 1 (whichused instantmessaging) and none in Study 3 reached an efficientoutcome. So, whereas Study 2 revealed that publiccommunication settings resulted in less outcome efficiency compared to private communication, Studies 1and 3 showed that inefficient outcomes are highlylikely when secret communication is allowed. This suggests that outcome efficiency is influenced not only bythe communication channels (face-to-face versus computermediated) being used (as shown inStudies 1 and2) but also by contextual factors such as whether public,private, or secret communication is possible.

DiscussionSummary of Findings

In this article, we experimentally investigated thefollowing theory of coalition micro-dynamics. First,the expectation of which coalitions (or proto-coalitions)

may form in the future if the current negotiation failscreates incentives for coalition members to extractfavorable agreements from their current bargainingcounterparts. These incentives can be nested andcircular. That is, an agreement within coalition A maydepend on the expected agreement in coalition B,which may depend on the expected agreement withincoalition C, and so forth. In other words, the natureand efficiency of any current agreement depends onthe shared expectations about future agreements.

Second, such future agreements not only may beless favorable to current coalition partners, but mostimportantly, future coalitions may consist of different

parties, relegating at least some of the current coalition members to the much less desirable role ofopposition party. This is a defining characteristic of

majoritarian bargaining: it is not necessary to reachconsensus among all negotiating parties to come toan implementable agreement.


Third, a "fear of exclusion" not only sustains currentcoalitions as equilibria but may also lead negotiatingparties to accept inefficient outcomes out of the fearthat the current coalition will be replaced by a newone which leaves them out of the final deal.

Fourth, to resolve the problem of exclusion, anadvantageous strategy for a player is to form a trustedbond with some of the other players and then jointlytake advantage of the weak bargaining position of theplayers not included. This approach is complicatedby the fact that the third player can always makea sufficiently attractive offer to induce one of themembers of the proto-coalition to defect from thepreliminary agreement. Indeed, maintaining the current proto-coalition while negotiating with the thirdparty over the remaining surplus constitutes the coredifficulty of this bargaining problem. Inefficient outcomes may result if players are "too concerned" with

maintaining the current agreement to successfullycome to an agreement with the third party. The resultis an inefficient outcome: money will be left on thetable. Moreover, the mere concern that the other party

may defect may lead a party to abandon the currentproto-coalition.

Fifth, the ability to form trusted bonds betweennegotiation parties will be influenced by various factors.The literature on communication in group decision

making suggests that the communication environmentin particular plays an important role in building thesebonds. For example, we found that both the communication channel (e.g., is nonverbal communicationpossible?) and the communication context (e.g., isthere an ability to communicate secretly?) influencedparties' intentions to trust others.We systematically varied the communicationchannels negotiating parties used to examine theinfluence of the ability to communicate publicly, privately, and secretly on the efficiency of negotiatedoutcomes. Across three studies, we found that outcome efficiency inmultiparty coalition negotiation isa function of the available communication channelsand the setting of the communication process itself.

More specifically, we found that face-to-face negotiations led consistently to more efficient outcomesthan CMC negotiations. Similarly, groups who wereforced to communicate publicly were able to reachmore efficient outcomes than groups with opportunities to communicate privately. Allowing participantsto communicate secretly further decreased outcomeefficiency until almost all negotiations involved two

way deals. Moreover, we found a similar pattern forcoalition dynamics in Studies 1 and 2.


16/19


In addition to these results, our approach yieldedtwo additional findings. First, we found evidence thatthe physical presence of the left-out party matters forcoalition outcomes. Proto-coalition members appearbe embarrassed to "shut the door" on the third partypermanently, as this may be considered sociallyunacceptable behavior. These findings are of interestbeyond the specific domain of our study, as they indicate how psychological factors and do-no-harm norms(J. Baron 1995) mitigate strategic incentives in political decision making.

Second, among the various versions of sequentialbargaining models of coalition formation, our resultsshowed strong evidence for the proto-coalitionapproach (D. P. Baron and Diermeier 2001). If giventhe opportunity, parties will take advantage of theopportunity to form "agreements to talk" that frequently exclude third parties. Negotiations on how tosplit the available pie are then conducted within suchproto-coalitions. This tendency is amplified whenplayers communicate in online settings.Our original intent in the study was to assess theability of the sequential bargaining models to accountfor the micro-dynamics of coalition negotiations. Wereasoned that the ability to maintain mutual trust inthe preliminary agreement would be a critical variablein explaining the ability to expand to three-playercoalitions. Then we conjectured that the ability tomaintain trust would be influenced by the communication channels available to the negotiators. In other

words, rather than asking people whether they trustedthe other party, a very difficult design and measurement challenge, we experimentally manipulated thedecision context that is expected to enhance or undermine trust. To be sure, our approach depends on thesetwo steps to hold, and it is possible to consider otherpossible explanations for the varying coalition behaviorthat do not involve trust. Nevertheless, the presentedevidence provides strong evidence for an approachbased on proto-coalitions.

Other DomainsWhile our focus in this study was on coalition micro

dynamics, our strategy to experimentally manipulatethe communication context has provided insights thatshould be of interest beyond the coalition formationcontext. Specifically, our results point to the importance of two independent communication effects ongroup decision making in other domains.

International negotiations over peace treaties orthe Doha Round of the World Trade Organization

(WTO) are other possible applications. In the case ofthe WTO negotiations, for instance, all members oftheWTO needed to agree on changes to the mutuallybinding existing trading regime; yet, if no global agreement is possible, subsets of the countries involvedcan choose to adopt regional or limited multilateralagreements instead.

In the domestic regulatory context, a case thatdemonstrates the complexity of multiparty negotiations is the 1997 national tobacco settlement negotiations between the six largest U.S. tobacco companies.Negotiations included attorney generals from morethan twenty states, a group of trial lawyers includingthe so-called "Casta?o Group," who were representing smokers in the class-action lawsuit Casta?o v.

American Tobacco Company, and the antismokingadvocacy group National Center for Tobacco-FreeKids (e.g.,Derthick 2002; Pertschuk 2001). Other keyplayers (althoughnot formally partof thenegotiations)included the Food and Drug Administration (FDA,under David Kessler) and antismoking advocates suchas C. Everett Koop, the former U.S. surgeon general,and Stanton Glantz, a university professor and keyfigure in California's Non-Smoker Rights movement.The eventual agreement included all six tobacco manufacturers and most of the other parties. Minnesota'sattorney general, Hubert H. Humphrey III, however,refused to join the agreement and eventually went tocourt. This lowered the value of the agreement to thetobacco companies because it only partially shieldedthem from legal liability. Moreover, since the settlementneeded an act of Congress, the larger the number ofparties who formally endorsed the agreement, themore likely itwould be enacted into law. Eventually,the settlement died in Congress, in large part becauseof vocal opposition to the agreement by Kessler, Koop,and Glantz.23 In 1998, the attorneys general and thetobacco companies agreed to a scaled-down agreement,theMaster Settlement Agreement (Derthick, 2002).Although the agreement was less comprehensive,24 italso required a far smaller coalition to be enacted.

Examining the impact of computer-mediated communication is also important for decision making withinorganizations, whether governmental agencies, parties,or interest groups (e.g., many organizations rely oncomputer-mediated technologies in decision making,although their serious use in political decision making isstill limited; see, for example, Carnevale, Pruitt, andSeilheimer 1981;Kahai andCooper 1999;Lewis andFry 1977;Lim andBenbasat 1993;Overly 1999;Poole,Shannon, and DeSanctis 1992; Swaab et al. 2002;

Valley, Moag, and Bazerman 1998; Wichman 1970),


17/19


because it provides amechanism for disentangling different dimensions of the communication context. InCMC, we can more precisely determine the communication options parties have available?for example,whether they are able to send each other private messages during the negotiation and whether the content ofthese messages or the fact that they are sent is commonknowledge. That is, rather than asking individuals whataspects of the situation mattered (e.g., certain visualcues or the extent to which they believed they couldmonitor the situation), we were able tomanipulate theirdecision environment to assess the importance of external factors on negotiation outcomes.

Also, our findings argue that in isolating processesunderlying outcome efficiency, examining within mediadifferences may be more adequate and precise thanexamining between media differences. In other words,rather thanmaking comparisons between the impact of avariety of channels such as telephone and e-mail on decision-making outcomes, we chose to focus on one specific aspect underlying human communication behavior(i.e., the publicness of communication). Although thisrestricts our ability to generalize our findings to the useof other media, this approach does control for a range ofother factors that typically disturb findings in comparative research (e.g., the influence of synchronicity, textualcommunication, visual and audio cues). Furthermore, itcan be argued that our studies were constrained by a onetime interaction with a set number of negotiating parties,potentially threatening our external validity. Indeed,work byMannix (1994) suggests thatanticipatingfutureinteraction has an important effect on the current negotiation, and future research may thus take such factors intocloser consideration.

Taken together, the results of three studies provideevidence for the idea that coalition bargaining undermajority rule can lead to inefficiencies as negotiatorsbecome unable to commit to preliminary agreements.Moreover, the research shows that the mutual trustneeded to establish stable agreements is stronglyinfluenced by the communication channel that is usedas well as the opportunity to communicate in public ornot. These insights further generalize to other politicaldomains and provide a deeper understanding on therole of communication in collective decision making.

Notes1. See Diermeier (2006) for a detailed account of these

methodological developments.2. A variant of this setup allows (nested) amendments to aproposal before it is voted on. This is the case of an open amend

ment rule (D. P. Baron and Ferejohn 1989).

3. In subsequent years, Baron systematically applied the modelto various aspects of government formation, such as different votingand proposal rules (Baron 1989) or parties with spatial preferences (Baron 1993).4. See Diermeier andMerlo (2000) for an empirical investigation of formateur selection rules.

5. See D. P. Baron and Diermeier (2001) and Baron,Diermeier, and Fong (2007) for models of strategic voting underproportional representation.

6. While these insights reappear in other forms of dynamiccoalition bargainingmodels (tobe discussed below), they aremosteasily illustrated in the context of the Baron-Ferejohn model.

7. The terms impatience and concern, of course, have psychological connotations. At this point, we use them loosely toidentify rational incentives, but we will later introduce a richer

psychological framework.8. See also Diermeier, Eraslan, and Merlo (2003). The origi

nal motivation for developing this alternative framework wasempirical. Despite repeated efforts, models based on theBaron-Ferejohn framework were unable to account for the existence of minority and super-majority governments.

9. Diermeier, Eraslan, and Merlo (2003) developed a structural estimation technique to incorporate these insights into theempirical study of coalition government.10.We used amodified version of the exercise published bytheKellogg Dispute Resolution Research Center, Northwestern

University. This particular contextualization was chosen to ensurepotential comparability with other studies using the same exercise. The form of thepay-off matrix is attributed toRaiffa (1982).

11. We will use a contextualized version of this problem in ourexperiments. Table 1 represents our experiment's characteristicfunction.

12. Ananalogous argument

holds for the otherproto-coalitions,A-C and B-C. It illustrates the fact that this characteristic func

tion has an empty core.13. Note that in our experiments, we only specify the charac

teristic function of the bargaining problem. We do not presupposea specific bargaining protocol. That is, rather than forming protocoalitions, negotiators may directly propose fully specified offers,as in Baron-Ferejohn bargaining, or they may exchange demands,as in the framework proposed by Morelli (1999).

14. Of course, if parties have the ability to adjourn the meeting, various other communication channels may be used,including phone calls, secret meetings, and so forth.

Negotiations with adjournments are difficult to replicate in alaboratory setting. However, as we will see below, computer

mediated communication allows us to faithfully implementsecret communication.

15. The software was developed at Northwestern Universityby Daniel Diermeier, Timothy J. Feddersen, and members ofNorthwestern's Academic Technology group. The AyeWare Website can be found at http://ayesite.northwestern.edu/ayeware2/index.html. A power point presentation describing the capabilities of Aye Ware and its use in a classroom environment can befound at http://ayesite.northwestern.edu/ayeware2/presentation

may-2004.ppt. A live demonstration and overview of the softwarecan be viewed on streaming video at http://2east.northwestern.edu/pres6.html.

16. Pretesting indicated that this was sufficient time to prepareand negotiate.17. There were no cases in which a coalition failed to fullyallocate its available amount.


18/19


18. We also ran the analysis for face-to-face versus CMC,excluding the one group that reached an impasse. The statisticalsignificance of the results did not change (%2(1)= 10.22;p < .001).

19. Participants were not allowed to use BlackBerries or cellphones (for text messaging).

20. Rubinstein's telling example is the "e-mail game" inwhich there is some (small) probability that amessage will belost. The classic example is two generals who need to coordinateon an attack and send messages back and forth. The key is thatthe generals will only attack if their strategic assessment of thesituation is common knowledge.

21. There were two impasses in the CMC pubUc-communicationcondition and three impasses in the CMC private-communicationcondition. There were no impasses in the face-to-face conditions.Excluding all impasses from the analysis had no statisticallysignificant effects.

22. This line of reasoning is explored in more detail in Swaab,Kern, Diermeier, and Medvec (in press).

23. For detailed accounts of the tortuous defeat of the settlement, see, for example, Derthick (2002) and Pertschuk (2001).

24. The agreement, for example, did not touch on the role ofthe FDA or protection from future liability.

ReferencesArkes, H. R., L. T. Herren, and A. M. Isen. 2000. The role of poten

tial loss in the influence of affect on risk-taking behavior.Organizational Behavior and Human Decision Processes42:191-93.

Arunachalam, V., and W. N. Dilla. 1992. Computer-mediatedcommunication and structured interaction in transfer pricingnegotiation. Journal of Information Systems 6:149-70.

Aumann, R. 1976. Agreeing to disagree. Annals of Statistics4:1236-39.

Aumann, R. 1995. Backward induction and common knowledgeof rationality. Games and Economic Behavior 8:6-19.

Axelrod, R. 1970. Conflict of interest. Chicago: Markham.Baltes, B. B., M. Dickson, M. P. Sherman, C. C. Bauer, and J. S.

LaGanke. 2002. Computer-mediated communication and groupdecision making: A meta-analysis. Organizational Behaviorand Human Decision Processes 87 (1): 156-79.

Baron, D. P. 1989. A Noncooperative Theory of LegislativeCoalitions. American Journal of Political Science 33:1048-84.

Baron, D. 1993. Government Formation and Endogenous Parties.American Political Science Review 87: 34-47.

Baron, D. P., and D. Diermeier. 2001. Elections, governments,and parliaments in proportional representation systems.

Quarterly Journal of Economics August: 933-67.Baron, D. P., D. Diermeier, and P. Fong. 2007. Policy dynamics

and inefficiency in a parliamentary democracy with proportionalrepresentation. Research Paper 1960, Stanford University

Graduate School of Business.Baron, D. P., and J. A. Ferejohn. 1989. Bargaining in legislatures.

The American Political Science Review 83 (4): 1181-1206.Baron, J. 1995. Blind justice: Fairness to groups and the do-no-harm

principle. Journal of Behavioral Decision Making 8:71-83.Carlson, J. R., and R. W. Zmud. 1999. Channel expansion theory

and the experiential nature of media richness perceptions.Academy of Management Journal 42 (2): 153-70.

Carnevale, P. J., D. J. Pruitt, and S. Seilheimer. 1981. Lookingand competing: Accountability and visual access in integrative bargaining. Journal of Personality and Social Psychology40:111-20.

Croson, R. T. 1999. Look at me when you say that: An electronicnegotiation simulation. Simulation and Gaming 30:23-37.Daft, R. L., and R. Lengel. 1986. Organizational informationrequirements, media richness and structural design. Manage

ment Science 32:554-71.Derthick, M. A. 2002. Up in smoke: From legislation to litigation

in tobacco politics. Washington, DC: CQ Press.Diermeier, D. 2006. Coalition government. In Oxford Handbook

of Political Economy, ed. B. Weingast and D. Wittman. NewYork: Oxford University Press.

Diermeier, D. and S. Gailmard. 2006. Self-interest, Inequality, andEntitlement in Majoritarian Decision-Making. QuarterlyJournal of Political Science 1(4): 327-350.

Diermeier, D., and A. Merlo. 2000. Government turnover in parliamentary democracies. Journal of Economic Theory 94:46-79.

Diermeier, D., H. Eraslan, and A. Merlo. 2003. A structuralmodel of government formation. Econometrica 71 (1): 27-70.

Diermeier, D., and R. B. Myerson. 1994. Bargaining, Veto Power,and Legislative Committees. CMSEMS working paper No.1089. Northwestern University.

Diermeier, D., and R. B. Myerson. 1999. Bicameralism and itsconsequences for the internal organization of legislatures.American Economic Review 89 (5): 1182-96.

Eraslan, H., and A. Merlo. 2002. Majority rule in a stochasticmodel of bargaining. Journal of Economic Theory 10:31-48.

Frechette, G., J. H. Kagel, and M. Morelli. 2005. Behavioralidentification in coalition bargaining: An experimental analysisof demand bargaining and alternating offers. Econometrica 73(6): 1893-1937.

Fulk, J. 1993. Social construction of communication technology.Academy ofManagement Journal 36 (5): 921-50.

Kahai, S. S., and R. B. Cooper. 1999. The effect of computermediated communication on agreement and acceptance.Journal of Management Information Systems 16 (1): 165-88.

Kahneman, D., and A. Tversky. 1979. Prospect theory: An analysisof choice under risk. Econometrica 47 (2): 263-91.

Kiesler, S., J. Siegel, and T. W. McGuire. 1984. Social psychological aspects of computer-mediated communication. American

Psychologist 39:1123-34.King, D. C, and J. C. Glidewell. 1980. Dyadic bargaining outcomes

under individualistic and competitive orientations. HumanRelations 33 (11): 781-803.Laver, M., and N. Schofield. 1990. Multi-party government: The pol

itics of coalition inEurope. Oxford, UK: Oxford University Press.Lewis, S., and W Fry. 1977. Effects of visual access and orien

tation on discovery of integrative bargaining. OrganizationalBehavior and Human Performance 20 (1): 75-92.

Lim, L. H., and I. Benbasat. 1993. A theoretical perspective ofnegotiation support systems. Journal of ManagementInformation Systems 9 (3): 27-44.

Loyd, D. L., M. C. Kern, and L. Thompson. 2005. Classroomresearch: Bridging the ivory divide. Academy of ManagementLearning & Education 4 (1): 8-21.

Madigan, D. 1996. Social Services. Evanstan, IL: KelloggDispute Resolution Research Center, NorthwesternUniversity.


19/19


Mannix, E. A. 1994. Will we meet again? The effects of power,distribution norms, and the scope of future interaction in smallgroup negotiation. International Journal of Conflict Management 5:343-68.

Mennecke, B. E., J. S. Valacich, and B. C. Wheeler. 2000. Theeffects of media and task on user performance: A test of thetask-media fit hypothesis. Group Decision and Negotiation9 (6): 507-29.

Morelli, M. 1999. Demand Competition and Policy Compromisein Legislative Bargaining, American Political Science Review93(4): 809-820.

Morley, I. E., and G. M. Stephenson. 1969. Interpersonal andinterparty exchange, a laboratory simulation of a negotiation at the plant level. British Journal of Psychology 60:543-45.

-. 1970. Formality in experimental negotiations: A validation study.British Journal of Psychology 61 (3): 383-84.Morris, M., J. Nadler, T. Kurtzberg, and L. Thompson. 2002.Schmooze or lose: Social friction and lubrication in e-mail

negotiations. Group Dynamics 6(1): 89-100.Morris, S., and H. S. Shin. 2003. Global Games: Theory and

Applications. In Advances in Economics and Econometrics(Proceedings of theEighth World Congress of theEconometric

Society), eds. M. Dewatripont, L. Hansen and S. Turnovsky.Cambridge, England: Cambridge University Press.

Naquin, C. E., and G. D. Paulson. 2003. Online bargaining and interpersonal trust.Journal ofApplied Psychology 88 (1): 113-20.

Overly, M. R. 1999. E-policy. New York: Amacom.Pertschuk, M. 2001. Smoke in their eyes: Lessons in movement

leadership from the tobacco wars. Nashville: VanderbiltUniversity Press.

Poole, M. S., D. L. Shannon, and G. DeSanctis. 1992.Communication media and negotiation processes. In Communication and negotiation, ed. L. L. Putnam and M. E. Roloff.Newbury Park, CA: Sage.

Purdy, J. M., and P. Nye. 2000. The impact of communicationmedia on negotiation outcomes. International Journal ofConflict Management 11 (2): 162-87.

Rangaswamy, A., and G. Shell. 1997. Using computers to realizejoint gains in negotiations: Toward an "electronic bargainingtable." Management Science 43 (8): 1147-63.

Rubinstein, A. 1982. The electronic mail game: Strategic behaviorunder almost common knowledge. American Economic Review79:385-91.

Rutter, D. 1987. Communication by telephone. Elmsford, NY:Pergamon.

Short, J.A. 1974. The effect of medium communication on experimental negotiation. Human Relations 27:225-34.

Short, J.A., E. Williams, and B. Christie. 1976. The social psychology of telecommunications. New York: John Wiley & Sons.

Suh, K. S. 1999. Impact of communication medium on task performance and satisfaction: An examination of media-richnesstheory. Information & Management 35:295-312.

Swaab, R., M. Kern, D. Diermier, and V. Medvec. In press. Whosays what to whom? Effects of communicator- and contentawareness on coalition formation. Social Cognition.

Swaab, R. I., V. Medvec, and D. Diermeier. n.d. Communicationmedia and negotiations: Meta-analyses on processes and outcomes. Under review.

Swaab, R. I., T. Postines, P. Neijens, M. H. Kiers, and A. C. M.Dumay. 2002. Multi-party negotiation support: The role of visualization's influence on the development of shared mental models. Journal of Management Information Systems 19 (1): 129-50.

Swaab, R. I., T. Postmes, and R. Spears. In press. Identity formation in multiparty negotiation. British Journal of Social

Psychology.Valley, K. L., J.Moag, and M. Bazerman. 1998. "Amatter of trust":Effects of communication on the efficiency and distribution of

outcomes. Journal of Economic Behavior & Organization34:211-38.

Walther, J. 1996. Computer-mediated communication: Impersonal,interpersonal, and hyperpersonal interaction. Communication

Research 23:3-43.Wichman, H. 1970. Effects of isolation and communication on

cooperation in a 2-person game. Journal of Personality andSocial Psychology 16(1): 114.

Documents

2008 Micro-Dynamics of Coalition Formation