The effects of groups' variety and disparity on groups' cognitive complexity

The Effects of Groups’ Variety and Disparity on Groups’ CognitiveComplexity

Petru Lucian CurseuTilburg University

Sandra SchruijerUniversity of Utrecht

Smaranda Boros“Babes-Bolyai” University Cluj-Napoca

This study examined the influence of group diversity conceptualized as disparity and asvariety on group cognitive complexity. Data on individual cognitive complexity andgroup cognitive complexity were collected in 44 groups using a conceptual mappingtechnique. Also data on the quality of teamwork processes and satisfaction werecollected using an individual questionnaire. The results indicate that (a) gender varietyhas a positive impact on group cognitive complexity, (b) cognitive disparity has anegative impact on group cognitive complexity, and (c) groups with a high averageindividual cognitive complexity have the highest cognitive complexity as a group onlyif the quality of their interactions is high.

Keywords: cognitive mapping, cognitive complexity, group dynamics

In modern organizations, complex cognitivetasks ranging from decision making to strategydevelopment are generally given to groupsrather than individuals (Cooke, Kiekel, Salas,Stout, Bowers, & Cannon Bowers, 2003;Devine, 2002; Weber & Donahue, 2001). It isbelieved that groups can use a greater pool ofknowledge to tackle complex information-processing tasks and be more effective thanindividuals. Therefore, the diversity in knowl-edge resources is a strategic advantage. How-ever, whether “n” different heads are alwaysbetter than “n” similar ones is still a matter ofdebate. This issue has been extensively investi-gated in the group diversity literature. It is ar-gued that in order for a team to be effective ithas to successfully integrate the individualknowledge of its members. This process was

labeled as knowledge integration by Okhuysenand Eisenhardt (2002), as the elaboration oftask information by Van Knippenberg, DeDreu, and Homan (2004), or as group cognitivecomplexity by Curseu (2006).

In this article, we will further on use the termgroup cognitive complexity to define the rich-ness of the collective knowledge structures thatemerge as a team-level phenomenon from theintegration of individual specialized knowledgethrough interpersonal interactions (Curseu,2006). Group cognitive complexity is an essen-tial characteristic of group cognition. Althoughmost of group cognition scholars agree thatgroup cognition is a group-level phenomenonthat emerges from the interplay between theindividual cognitions of group members andtheir interactions (Rentsch & Woehr, 2004;Salas & Fiore, 2004), there is little emphasis inthe literature on the emergence of group cogni-tion as a group-level phenomenon (Curseu,2006).

Also, in the diversity literature questions con-cerning what types of attributes are more likelyto be beneficial for group cognitive complexityor under which conditions a diverse group caneffectively use the variety of knowledge andexpertise of its members, remain largely unan-

Petru Lucian Curseu, Department of Organisation Stud-ies, Tilburg University; Sandra Schruijer, Utrecht School ofGovernance, University of Utrecht; and Smaranda Boros,Department of Psychology, “Babes-Bolyai” UniversityCluj-Napoca.

Correspondence concerning this article should be ad-dressed to Petru Lucian Curseu, Department of Organiza-tion Studies, Tilburg University, Room S161, Warande-laan 2, P.O. Box 90153, 5000 LE Tilburg, The Netherlands.E-mail: [email protected]

Group Dynamics: Theory, Research, and Practice Copyright 2007 by the American Psychological Association2007, Vol. 11, No. 3, 187–206 1089-2699/07/$12.00 DOI: 10.1037/1089-2699.11.3.187

187

swered (Milliken & Martins, 1996; Williams &O’Reilly, 1998). Several taxonomic approacheshave been used to make sense of the puzzlingeffects of group diversity on group performance(Pelled, 1996; Harrison & Klein, 2007) startingfrom the assumption that different types of di-versity yield different effects on performance.Harrison and Klein (2007) argue that groupdiversity can be operationalized as separation(e.g., differences in opinions among the groupmembers), variety (e.g., differences in types ofknowledge or experience among the groupmembers), or disparity (e.g., differences in as-sets or resources among the group members),and only variety will foster a higher pool ofknowledge within a group. This taxonomic ap-proach has good potential for explaining themixed effects of group diversity on group per-formance because on the one hand it accommo-dates for other existing taxonomies (e.g., Pelled,1996; Milliken & Martins, 1996) and on theother hand it connects each diversity type withthe underlying mechanisms of group perfor-mance (Harrison & Klein, 2007).

The first aim of this article is to explore therole of group interaction processes on groups’cognitive complexity. We argue that the qualityof interpersonal interactions in a group is centralfor the integration of individual knowledgestructures (individual cognition) into a groupknowledge structure (group cognition) and thatthe quality of teamwork moderates the relationbetween the average individual cognitive com-plexity and group cognitive complexity. Thesecond aim is to use the taxonomy introducedby Harrison and Klein (2007) to explore thedifferential impact of group disparity and vari-ety on group cognitive complexity.

We extend existing research in several ways.First, the article contributes to the group diver-sity debates and provides an empirical test forthe theoretical propositions raised by the dis-tinction between disparity and variety as typesof group diversity. According to these proposi-tions, cognitive disparity is illustrative for thevertical differentiation within groups and is ex-pected to have a negative impact, while varietyis illustrative for a horizontal differentiationwithin groups and has a positive impact ongroup cognitive complexity. Second, the articleadds to the group cognition literature by explor-ing the role of interpersonal interactions withina group on the emergence of group cognition

(operationalized as a group’s cognitive com-plexity). Finally, the article sheds some light onthe impact of group disparity on group mem-bers’ satisfaction. We will start by shortly re-viewing the literature on group cognitive com-plexity as well as on group diversity with anemphasis on cognitive diversity.

Cognitive Complexity in Groups and theMeaning of Collaboration

The concept of cognitive complexity (CC)has initially been introduced by Bieri (1955) asa personality trait and has subsequently beenresearched within the framework of personalconstruct theories (Kelly, 1955). Later it hasbeen redefined as a characteristic of informationprocessing in cognitive systems (Schroder,Driver, & Streufert, 1967). CC refers to thecomplexity of the knowledge structures in acognitive system, and it describes the sophisti-cation of those cognitive structures that are usedfor organizing and storing cognitive contents(Kelly, 1955; Goodwin, Wofford, & Harrison,1990; Curseu & Rus, 2005). High CC reflects aflexible and adaptive orientation in informationprocessing (Schroder et al., 1967).

Groups are sociocognitive systems (Curseu,2003; Hinsz, Tindale, & Vollrath, 1997; Hollan,Hutchins, & Kirsch, 2000; Hutchins, 1995).They develop, store, and use cognitive repre-sentations (Curseu, 2003). Because cognitivesystems can vary in complexity on a continuumranging from cognitive simplicity to cognitivecomplexity (Schroder et al., 1967), the conceptof cognitive complexity can also be applied togroups as a characteristic of group cognition.According to the group cognition approaches(Curseu, 2003; Hinsz, Tindale, & Vollrath, 1997;Hollan et al., 2000; Hutchins, 1995; Mohammed,Klimoski, & Rentsch, 2000; Rentsch & Woehr,2004; Weick & Roberts, 1993), group cognitiondepends on the knowledge of individual groupmembers as well as on their interaction processes.

One empirical study (Hendrick, 1979) sug-gests that the individual cognitive complexity ofgroup members has a positive effect on groupprocesses and performance. He found thatgroups composed of cognitively complex indi-viduals used information cues more efficiently,interacted more easily, and came up with cor-rect solutions for a puzzle faster than groupswhose members were low on cognitive com-

188 CURSEU, SCHRUIJER, AND BOROS

plexity. Nonetheless, Hendrick (1979) did notinvestigate the cognitive complexity of thegroup as a whole and did not take into accountthe interactions among group members. Thecognitive complexity was measured only at theindividual level, not at the group level. Only hisinferences regarding performance pertain to thegroup level. By ignoring the interactions be-tween group members, Hendrick (1979) did notconceptualize the group as a social system butrather as an addition of individuals varying incognitive complexity. In real groups interper-sonal interactions play a crucial role in the de-velopment of group cognition, and to date thereis little empirical evidence for the interplay be-tween teamwork processes and the cognitivecomplexity of the group members in generatinggroup cognition (Curseu & Rus, 2005).

Although group cognitive complexity waspreviously discussed in relation to group perfor-mance—as the elaboration of task relevant in-formation (Van Knippenberg et al., 2004)or knowledge integration (Okhuysen &Eisenhardt, 2002)—no clear operationalizationand measurement of the these concepts as agroup-level phenomenon was developed. In thepresent study, group cognitive complexity isoperationalized as the number of independentconcepts used by the group to define (represent)a particular situation or knowledge domain andthe number and variety of connections amongthese concepts (Curseu & Rus, 2005). This op-erationalization is consistent with the definitionof cognitive complexity in cognitive systems ingeneral. In a cognitive system characterized byhigh cognitive complexity, information process-ing is defined by the use of many constructswith many relations among them (Schroder etal., 1967). The cognitive complexity of thegroup is domain specific: groups can be cogni-tively complex in some areas and cognitivelysimple in others. Therefore, group cognitivecomplexity describes the richness of the knowl-edge representations held by the group for aparticular knowledge domain.

One possibility for groups to represent de-clarative knowledge is through natural lan-guage. Conceptual networks are therefore cen-tral representation forms. A conceptual networkis a graphical representation, which consists ofseveral interconnected concepts that the groupuses to make sense of a task. Although variousapproaches have been used to represent group

cognition, the most frequently reported tech-nique has been cognitive mapping (Axelrod,1976; Bougon, 1983; Huff, 1990). The com-plexity of the group’s cognitive map is illustra-tive for the cognitive complexity of the group.

Most of the studies use aggregation methodsto combine individual cognitions into sharedmental models (Cooke, Salas, Cannon-Bowers,& Stout, 2000; Cooke et al., 2004; Lagan-Fox,Code, & Langfield-Smith, 2000; Mohammed,Klimoski, & Rentsch, 2000). However, as ar-gued, group cognition is a group-level phenom-enon, and any evaluation method must (1) ad-dress the group as a whole, (2) demonstrategroup members’ agreement with regard to theevaluated construct, (3) demonstrate that theresults discriminate across groups, and (4) re-flect group interaction processes (see for moredetails Bar-Tal, 1990). A simple aggregation ofindividual mental models does not satisfy allfour criteria. However, cognitive mapping canbe adapted to meet the four criteria. A card-sorting technique is suitable for such a purpose.Relevant concepts for the knowledge domain ofthe group (to be elicited by interview, documentanalysis, verbal protocol analysis—see for de-tails Carley, 1993, 1997; Mohammed et al.,2000) are to be written on independent cards,and then the group members are instructed toorganize them in a way that makes sense to thegroup as a whole. This method can be used tostudy the structure of the conceptual networksdeveloped by groups at the group level of anal-ysis.

In this study we use such a modified cogni-tive mapping technique to understand how in-dividuals and groups make sense of the conceptof “collaboration.” Thus, we elicit conceptualnetworks about the phenomenon of collabora-tion. When it is used as a group task, the groupmembers have to reach consensus on the struc-ture of the conceptual network that best repre-sents the group as a whole (Curseu, 2003;Curseu & Rus, 2005). Although this consensusbased method reflects the group as a whole (itsatisfies all the requirements stated by Bar-Tal,1990, for valid holistic evaluation methods), wehave to acknowledge, in line with Sundstrom,Busby, and Bobrow (1997), the fact that reach-ing consensus in this particular task does notnecessarily mean complete unanimity (the tech-nique might involve compromises among thegroup members). We will further address the

189VARIETY AND DISPARITY ON COGNITIVE COMPLEXITY

issue of computing the complexity of cognitivemaps in the section on procedure.

Group Diversity

Group diversity refers to the degree of differ-entiation that exists among the group memberswith respect to a particular attribute (Harrison &Klein, 2007). Although during the last de-cades, group diversity received a considerableamount of interest (Weber & Donahue, 2001;van Knippenberg et al., 2004), scholars stilldisagree if group diversity is beneficial or notfor group performance (Milliken & Martins,1996; Williams & O’Reilly, 1998). It is gener-ally believed that heterogeneous groups aremore creative and reach better decisions, yetexperience more difficult group interaction pro-cesses (suboptimal communication, conflict,stereotyping) than homogeneous groups. How-ever, homogeneous groups solve problemsmore quickly, are more cohesive, and groupmembers are more satisfied with their coopera-tion (Milliken & Martins, 1996; Williams &O’Reilly, 1998; Weber & Donahue, 2001).

Research on group diversity addressed a largevariety of attributes, from demographic (e.g.,age, gender) to cognitive ones (e.g., attitudes,values), and currently several diversity at-tributes taxonomies exist. The most used onesdistinguish between visible versus less visibleattributes (Milliken & Martins, 1996) and be-tween highly job-related versus less job-relatedattributes (Pelled, 1996). The first taxonomy(visible vs. less visible attributes) emerged fromreviews of the group diversity literature, and itdoes not describe the mechanisms that connectgroup diversity with performance, while thesecond was developed starting from mecha-nisms that could explain the differential impactof diversity on performance, yet it received littleto no empirical support (Weber & Donahue,2001).

More recently, Harrison and Klein (2007)introduced a new distinction. They differentiatebetween group diversity as separation (differ-ences in beliefs, attitudes, and values), variety(differences in functional background and typeof expertise) and disparity (inequalities in sta-tus, power, and resource availability). Of allthese types, only variety is expected to have apositive influence on group effectiveness(Harrison & Klein, 2007). Separation refers to

differences in the lateral disposition of thegroup members on a continuum defined by acertain diversity trait. Separation reflects a bi-modal distribution, with half of group membersat the highest (e.g., they have strong similarreligious beliefs), the other at lowest endpointsof the considered variable’s continuum (e.g.,they have no religious beliefs). Variety refers tothe composition of differences in kind, sourceor category of relevant knowledge or experienceamong group members. It reflects a uniformdistribution with an even spread of membersacross all possible categories of a variable (e.g.,a group high on variety is a group composed ofpsychologists, sociologists and anthropolo-gists). Disparity refers to the composition of(vertical) differences in proportion of sociallyvalued assets or resources held among groupmembers, pointing to an inequality or relativeconcentration. Disparity reflects a positivelyskewed distribution with one member at thehighest point on the continuum (one memberthat has access to a particular form of resource,like money, knowledge, or expertise) of theconsidered variable, others at the lowest (mostof the members have no access to the previouslymentioned resources) (Harrison & Klein, 2007).The taxonomy introduced by Harrison andKlein (2007) opens new ways of looking atgroup diversity. The same attribute, gender forexample can be understood as variety (e.g., ifgroup members have different task relatedknowledge structures emerging from qualita-tively different gender-specific life experi-ences), disparity (e.g., if the most powerful per-son in a group of several women and one man isthe man), or separation (e.g., if all women in agroup are feminist and all men are misogynists).

The complexity of group cognition (group’scognitive complexity or elaboration of task rel-evant information) is the linking pin betweengroup diversity and group performance, or asCooke et al. (2003) stated, the cognitive under-pinnings of group performance. From the typesof diversity described by Harrison and Klein(2007) only variety has a positive impact on theelaboration of task relevant information, whiledisparity has disruptive effects.

Group variety is associated with a largerknowledge repertoire which will ultimately bereflected in the complexity of the cognitive mapproduced by the group. There is, however, therequirement that the variety of knowledge


group members have should be relevant for thetask (Schruijer & Vansina, 1997). This require-ment is also supported by the meta-analysis byBower, Pharmer, and Salas (2000) who showedthat the advantages of group heterogeneity arestrictly dependent on the task. The specific taskused in this study is to elicit cognitive mapsabout the concept of collaboration. Gender di-versity is certainly related to this particular taskand can be conceptualized as variety. The deci-sion quality is higher in gender mixed groupsdue to the qualitatively different contributionsto teamwork men and women have (Rogelberg& Rumery, 1996); therefore, it is expected thatwhen mapping collaboration men and womenwill contribute with different knowledge struc-tures rooted in different life experiences withcollaboration.

Group disparity on the other hand is morelikely to trigger misunderstandings and conflictsand will have a negative impact on the com-plexity of the cognitive map of the group. Fewattempts have been made to study the diversityof the groups in terms of the content and struc-ture of the cognitive representations held byindividual members. Knowledge and expertiseis certainly a relevant resource for groups, es-pecially when they have to perform cognitivetasks. Therefore, the complexity of members’understanding of the issue at hand is a keyaspect of group disparity. The disparity attributewe consider in this study is the complexity ofthe individual cognitive map group membershave about collaboration.

Hypotheses

The efficiency of groups (operationalized asthe time needed to complete a cognitive-mapping task) is generally less than that ofindividuals. However, we expect that whenmaking decisions or discussing a relevant issue(e.g., the meaning of collaboration), people withsimilar cognitive structures reach consensusfaster than groups consisting of people withdifferent cognitive structures. Based on this, it ispredicted that both group variety and disparitywill have a positive impact on the time neededto reach consensus within the group. However,under the conditions of high cognitive disparity(one group member with a complex knowledgestructure and the others with simple cognitivestructures), the conversion theory of minority

influence (Moscovici, 1980) suggests that mostprobably the informed or expert minority (inthis case the member with a complex knowl-edge structure) will lead the group toward adeeper information processing of the issue athand and a greater cognitive activity in thegroup. To reconcile the different perspectives,the group needs time and, as argued by Harrisonand Klein (2007), disparity is also likely to beassociated with process losses. Because of that,it is expected that group diversity as disparitywill have a stronger impact on the length ofgroup discussions (time to reach consensus)than group diversity as variety.

Hypothesis 1: Both disparity and varietywill have a positive impact on the length of timeneeded to reach consensus in groups with dis-parity having the strongest effect.

However, the criterion for effectiveness is notthe time needed to reach consensus, but thevolume of knowledge pooled during group dis-cussions. Groups of people with different back-grounds (high group variety) are expected toconstruct more complex conceptual networks orcognitive maps than homogeneous groups(Curseu, 2003; Harrison & Klein, 2007). Asargued in the previous section, group diversityas variety is expected to have a positive impacton the cognitive complexity of groups, in thesense that the pool of knowledge within thegroup will be higher in groups composed ofmembers with different types of experiencesand expertise (cognitively diverse groups),while group diversity as disparity will have theopposite effect on group cognitive complexity.Therefore, it is anticipated that group diversitywill only be beneficial for group cognitive com-plexity if it is reflected by the variety in per-spectives that are highly relevant for the task.

The attributes that ensure greater informationrichness and variety within groups may differ.Members with different educational back-grounds or with different types of expertise arelikely to use their differences in perspectives,ensuring a higher cognitive complexity of thegroup. Most of the studies concerning the ef-fects of demographic diversity dealt with indi-vidual level consequences (low satisfaction,high turnover; see for details, Sacco & Schmitt,2005). However, demographic diversity issometimes an accurate proxy for less visibledifferences (e.g., knowledge, experiences) thatare in fact relevant for the task. In a policy


development group for racial issues, for exam-ple, racial diversity will be closely associatedwith group variety, and due to a greater infor-mational richness, it will ultimately be benefi-cial for the group outcomes. A diverse groupwith respect to gender will also have a highercognitive complexity if the task requires a va-riety of knowledge that emerges from genderspecific experiences. A similar argument holdsfor collaboration. Gender is a task-relevant di-versity attribute because men and women havedifferent experiences with and attitudes towardcollaboration (see, e.g., the results concerninggender differences in negotiations, Stuhlmacher& Walters, 1999). In this respect, gender diver-sity can be conceptualized as variety and ex-pected to have a positive impact on a group’scognitive complexity.

Further, there is evidence showing that mixedgender groups generate a higher number of al-ternatives and that they discuss a higher numberof ideas (Schruijer & Mostert, 1997), and theyare more creative (Hoffman & Maier, 1961) ascompared to homogeneous groups. Also inmixed gender groups the quality of decisions ishigher than in homogeneous gender groups dueto the qualitatively different contributions toteamwork men and women have (Rogelberg &Rumery, 1996). Because the cognitive com-plexity of groups also depends on the number ofissues discussed, as well as on the quality ofteamwork, these results support the predictionthat gender diversity will be beneficial for thecognitive complexity of groups (in this case theconceptual network of collaboration). To sum-marize, according to previous research, genderdiversity (1) can be understood as variety, and(2) it is beneficial for group creativity; there-fore, our prediction is that in tasks that require avariety of informational resources that are oneway or another associated to gender-specifictraits, gender variety has a positive impact on agroup’s cognitive complexity.

Hypothesis 2: Gender variety has a posi-tive impact on the cognitive complexity ofgroups.

Group disparity, however, is expected tohave a different impact on a group’s cognitivecomplexity. Group diversity conceptualized asdisparity involves an asymmetry within a groupconcerning the distribution of valued assets,with few group members having a particularasset and the majority of group members not

possessing it (Harrison & Klein, 2007). If fewmembers within a group have a high level ofexpertise or, in a more general sense, a highlevel of cognitive complexity, while the major-ity of the group members have a low level ofcognitive complexity (high disparity), the cog-nitive complexity of the group will be also low.A similar argument holds for the diversity inindividual cognitive complexity (further re-ferred to as cognitive disparity). If only onemember in a group has a highly complex con-ceptual network for collaboration, it is verylikely that he or she will try to persuade theothers to accept his or her perspective on theissue (see, e.g., the research on minority influ-ence, Moscovici, 1980; Wood et al., 1994).Therefore, it is expected that the general com-plexity of the conceptual network developed bythe group will be rather low, reflecting mostlyone perspective on collaboration.

Hypothesis 3: Group cognitive disparityhas a negative impact on group cognitive com-plexity.

Cognitive disparity will most probably influ-ence the individual group members as well. Itseems reasonable to argue that such asymmetryin cognitive complexity will frustrate the mem-bers with a higher cognitive complexity; theywill perceive the individual contributions to thegroup task as unequal, and they will be lesssatisfied with group processes and outcomesthan members with a low cognitive complexity.

Hypothesis 4: Group members with ahigher CC than the group’s CC will be lesssatisfied about the quality of teamwork as com-pared with members with a lower individual CCthan the group’s CC.

Hypothesis 5: Participation on the task willbe perceived as less equal by group memberswith a higher individual CC than the group’sCC as compared with members with a lowerindividual CC than the group’s CC.

The individual cognitive complexity of thegroup members is expected to have a positiveinfluence on a group’s cognitive complexity. Inline with the empirical results reported byHendrick (1979), we argue here for a direct andpositive impact of average individual cognitivecomplexity on group’s cognitive complexity.The higher the cognitive complexity of individ-ual group members, the higher the group’s cog-nitive complexity will be. However, group cog-nitive complexity is more than just an aggrega-


tion of individual cognitive complexities. Groupinteractions (debates, negotiations, discussions)can shape and reshape the content and structureof the group’s conceptual network (or cognitivemap), and new patterns of connections amongconcepts might emerge (which cannot be cap-tured through a simple aggregation of individ-ual cognitive maps; Curseu, 2003; Curseu &Rus, 2005). It is therefore reasonable to arguethat groups with a high quality of interpersonalinteractions (further on referred to as high team-work quality—see for details Hoegl &Gemuenden, 2001) will benefit the most fromtheir members’ level of cognitive complexity.In conclusion, we predict that teamwork qualitywill have a moderating effect on the relationbetween the average individual cognitive com-plexity within a group and the group’s cognitivecomplexity.

Hypothesis 6: The teamwork quality moder-ates the relation between the average individualcognitive complexity and the group cognitivecomplexity, in the sense that the positive effectof average individual cognitive complexity ongroup cognitive complexity will be accentuatedin groups that experience a higher teamworkquality.

Method

Sample

A sample of 132 students (with an averageage of 20.16 years; 74 women and 58 men)participated in a cognitive mapping session inexchange for extra credits for a social psychol-ogy course. Only students with previous group-work experience were used as respondents inour sample. First individually and then ingroups of two, three, or four, the students wereasked to select from 40 concepts those thataccording to them are related to collaborationand subsequently to organize them in a way thatmakes sense to them. Students were grouped inhomogeneous (14 male groups and 15 femalegroups) and heterogeneous (15) groups. Afterrealizing the group maps, the participants wereasked to fill in a questionnaire evaluating satis-faction, equal participation to the group out-come, and teamwork quality (communication,collaboration, planning, organizing, conflict,and process efficiency).

Procedure

We used a card-sorting variant of a concep-tual mapping technique to explore the way inwhich individuals and groups represent collab-oration. Collaboration is a concept that suits thecontext of this study because the task contentrequires a variety of knowledge that is associ-ated with demographic attributes (e.g., gender-specific experiences with collaboration are arelevant source of variety for this particulartask). Another conjecture behind the use of thisconcept is that the concept of collaboration andits links to other relevant concepts can be rep-resented as a network. Students build a concep-tual network, in which the relevant concepts arerepresented as nodes and the relations amongthem are represented as lines (Bougon, 1983;Calori, Johnson, & Sarnin, 1994). This concep-tual mapping technique is particularly relevantfor our research because of its vast potential toencompass complex relations and interdepen-dencies in a conceptual domain.

In order to obtain the concepts to be used inthe conceptual mapping, we first interviewedfive students about their experiences with col-laboration. We then used a free associationtechnique to elicit the main concepts related tocollaboration from an independent sample of 80students. In this way we made sure that wefound the most relevant concepts used by thisparticular group (students) to define collabora-tion. From the interview and from the free as-sociation task, the 40 most important conceptswere selected. A list of the concepts rankedaccording to their frequency in the free associ-ation task is presented in Appendix A. Giventhe fact that the respondents had to map theconcepts twice and fill in two questionnaires,the time span of the study was of critical im-portance. Forty concepts can be mapped in areasonable amount of time (around 40 minutes),and in the same time they give a sufficientrichness of the conceptual domain to bemapped. Every concept was written on a differ-ent card. The respondents received an envelopewith the 40 concepts and an A3 blank sheet ofpaper and glue. Their task was to distribute theconcepts on the sheet in such a manner that theirspatial proximity would reflect the extent towhich they were related. Afterwards, they wereinstructed to draw the connections they seeamong the concepts and to specify the nature of


the relations between concepts. The respondentswere instructed to use only the concepts thataccording to them describe or are related tocollaboration; therefore, the decision on howmany concepts are relevant for collaborationand need to be included in the cognitive mapwas made by each individual or group.

The complexity of 132 individual and 44group cognitive maps was investigated usingthree indicators: (1) map connectivity, a countof the number of connections established be-tween the concepts, (2) map diversity, a countof the number of distinct types of relationsestablished between the concepts (a list anddescription of possible types of relations amongconcepts is presented in Appendix B—see fordetails also Gomez, Moreno, Pazos, & Sierra-Alonso, 2000), and (3) the number of conceptsused in the map. We computed the map com-plexity index based on the following formula:complexity � (connectivity*diversity)/numberof concepts (for details, see Curseu & Rus,2005). For an illustration on how cognitive mapcomplexity was computed see Appendix C.This formula reflects the embeddedness of theconcepts in the conceptual network at hand;therefore, it illustrates the richness of the con-ceptual knowledge structure developed bygroups about collaboration (as it is specified inthe definition of cognitive complexity). Previ-ous empirical research shows that the cognitivecomplexity of student groups (evaluated by thisformula) positively and significantly correlateswith the general performance of the group (thegroup grade for the project from which the con-cepts to be mapped were extracted), with the num-ber of ideas exchanged during group discus-sions, with the perceived effectiveness of thegroup as well as with the quality of teamwork(Curseu, 2003). Finally, as argued before, thecognitive mapping method meets the validitycriteria discussed by Bar-Tal (1990) for holisticevaluation methods in group research.

Questionnaire

Group members’ satisfaction with the group,perceived equal participation to the group out-come, and teamwork quality (collaboration, or-ganizing, conflict, and process efficiency) wereevaluated with an individual questionnaire com-pleted after the group cognitive mapping ses-sion.

Satisfaction was evaluated using two items(“How satisfied are you with the group pro-cess?” and “How satisfied are you with theoutcome of the group?”), rated on a 5-pointLikert scale (from 1 to 5). Cronbach’s alpha forthis scale is 0.83.

Equal participation to the group map wasevaluated with four items designed specially forthe cognitive mapping session (e.g., “The finalmap reflects the ideas expressed by all groupmembers,” “All group members participatedequally in creating the group cognitive map,”“My ideas are reflected in the final product ofthe group,” and “The ideas of all group mem-bers were incorporated into the final groupmap”), rated on a 5-point Likert scale (from 1 to5). Cronbach’s alpha for this scale is 0.79.

The quality of teamwork processes was eval-uated using four items (rated on a 5-point Likertscale, from 1 to 5) related to collaboration,organizing, conflict, and process efficiency. Pre-vious research emphasized the unitary factorstructure of the scales evaluating different facetsof teamwork (Curseu, 2003; Eby, Meade,Parisi, & Douthitt, 1999; Hoegl & Gemuenden,2001). The content of the items was developedusing examples from Eby, Meade, Parisi, andDouthitt (1999) and Hoegl and Gemuenden(2001). Consequently, the four items were usedtogether to evaluate teamwork quality. Cron-bach’s alpha for this scale is 0.82, with the itemevaluating conflict being reverse coded.

A principal components analysis of the scalerevealed a unitary factor structure (similar toprevious research Curseu, 2003; Eby, Meade,Parisi, & Douthitt, 1999; Hoegl & Gemuenden,2001), explaining 70% of the variance with thefollowing factor loadings: Conflict .70 (“Howoften did members of your group disagree orexpressed different opinions in the group?”—reverse coding), Collaboration .83 (“Groupmembers collaborated well and worked togetherinterdependently to achieve the task”), Organiz-ing .90 (“Group activities were well orga-nized”), and Process efficiency .86 (“The groupused effective processes to achieve the task”).

In order to justify aggregation of individualscores into group scores, we used the procedureintroduced by James, Demaree, and Wolf(1984) to estimate the interrater reliability (theindex of agreement within groups). The withingroup agreement index (Rwg) can take valuesbetween 0 and 1, and generally, a value of 0.70


or higher is considered to reflect a reasonableamount of agreement within a group (James,Demaree, & Wolf, 1984). Table 1 summarizesthe Rwg for each variable, with the maximum,minimum, mean, and the standard deviation ofthe Rwg scores.

After the within-group agreement was com-puted and verified, the individual scores of thegroup members were aggregated into groupscores by computing the group mean.

Group Diversity

The diversity index for gender was computedusing a formula proposed by Teachman (1980)and widely used in group diversity literature(Williams & Mean, 2004). The formula is:

H � � �i � 1

s

Pi(ln Pi)

where i represents a particular category, s is thetotal number of categories, and Pi is the propor-tion of the members belonging to the i category.If a group consists of members belonging to scategories, and Pi probability is assigned to agiven category, then the H index is a measure ofgroup heterogeneity (structural diversity). Thehigher the value of the index, the higher is thevariety of the group. The theoretical maximumfor H depends on the total number of categories(s; Williams & Mean, 2004). Since gender is adichotomous variable, we only had two catego-ries in our formula. For groups consisting ofonly one category, H � 0. This way of com-puting group diversity is consistent with theconceptualization of group diversity as variety(see for details, Harrison & Klein, 2007).

Cognitive disparity within groups was com-puted using the coefficient of variation. Accord-ing to Harrison and Klein (2007) the coefficient

of variation is a suitable method for computinggroup disparity. It is computed by dividing thestandard deviation with the group average cog-nitive complexity. It reaches a maximum whenn-1 group members are at the lower end of thescale, and one of the group members is at thehigher end of the scale (Harrison & Klein,2007). In addition, a heuristic method based onthe conceptualization of disparity by Harrisonand Klein (2007) was used to compare the re-sults. The groups were divided in three catego-ries: minimum, medium, and maximum dispar-ity. Four groups had minimum disparity (groupsin which the majority of the group members hada high cognitive complexity and groups inwhich all group members had approximatelythe same level of cognitive complexity). Twenty-seven groups had medium cognitive disparity(groups in which the individual cognitive com-plexity ranged from low to high and the differ-ences were compressed). Thirteen groups hadmaximum cognitive disparity (groups in whichonly one of the members had a very high cog-nitive complexity); see also Harrison & Klein,2007, Figure 1).

Results

The results of this study are reported at twolevels of analysis. Hypotheses 1, 2, 3, and 6concerned the group level of analysis, whileHypotheses 4 and 5 concerned variables evalu-ated at the individual level of analysis (satisfac-tion and equal participation). Table 2 shows themeans, standard deviations, and correlations forthe aggregated variables (group level of analy-sis), while Table 3 shows the means, standarddeviations, and correlations for the individuallevel of analysis.

To test Hypothesis 1, a stepwise regressionanalysis was performed, with gender varietyintroduced in the first step and cognitive dispar-ity introduced in the second step as predictorsfor the time needed to reach consensus in thecognitive mapping. The results of this regres-sion analysis are presented in Table 4. InModel 2, the standardized beta coefficient forcognitive disparity is positive and significant,while the standardized beta coefficient for gen-der variety is positive, but not significant. The Fchange when disparity is introduced in themodel is significant [F(1, 41) � 13.49, p �.001]; therefore, we can conclude that Hypoth-

Table 1Within Group Agreement Indices (Rwg)

N Min. Max. M SD

Rwg equal participationwithin groups 44 .75 1.00 .86 .06

Rwg quality of teamworkprocesses 44 .75 1.00 .84 .07

Rwg satisfaction 44 .75 1.00 .85 .08


esis 1 which predicts that both group disparityand variety will have a positive impact on thetime needed to reach consensus, and the effectof disparity will be the stronger of the two issupported.

A second regression analysis was conductedto test Hypotheses 2 and 3, and the results arepresented in Table 5. In this regression analysis,the dependent variable was the complexity ofthe group cognitive map, while the independentones are gender variety and cognitive disparity.Hypothesis 2 states that gender variety has apositive impact on group’s cognitive complex-ity, and as shown by the positive standardized

beta coefficient, it is marginally supported. Hy-pothesis 3 states that cognitive disparity has anegative impact on group’s cognitive complex-ity. This hypothesis also received only marginalsupport since the standardized beta coefficient isindeed negative, but only marginally signifi-cant.

In the operationalization of disparity,Harrison and Klein (2007) argue that the use ofa coefficient of variation to compute disparityinduces the effect of a particular moderatingstructure of within-group data and thereforethey recommend to use a modified regressionequation in which the hidden effect of the mean

Table 2Means, Standard Deviations, and Correlations for the Group Level of Analysis

M SD 1 2 3 4 5 6 7

1. Gender variety .21 .262. Cognitive disparity .35 .15 .023. Average ICC 2.88 .74 .22 �.214. Group CC 2.64 1.11 .21 �.25 .56**

5. Equal participationwithin groups 3.88 .51 �.13 �.28 �.11 .31*

6. Quality ofteamworkprocesses 3.92 .59 �.10 �.32* �.01 .42** .83**

7. Satisfaction 3.85 .55 .12 �.18 .04 .39** .79** .72**

8. Time to reachconsensus 23.36 6.39 .21 .49** .21 .17 �.01 �.10 .05

Note. ICC � individual cognitive complexity; CC � cognitive complexity; N � 44 groups.* p � .05. ** p � .01.

Table 3Means, Standard Deviations, and Correlations for the Individual Level of Analysis

M SD 1 2 3 4 5 6 7

1. Age 20.16 1.952. Gender 1.56 .49 .103. Equal participation

within groups 3.86 .64 �.01 .084. Quality of

teamworkprocesses 3.90 .72 .01 .05 .71**

5. Satisfaction 3.84 .76 �.04 .16 .63** .64**

6. Time to make theindividualcognitive map 33.94 7.58 �.06 .03 �.13 �.18* �.24**

7. ICC 2.89 1.17 �.03 .04 �.02 .07 .03 .028. Difference

between individualand group CC .25 1.29 �.01 �.01 �.24** �.24** �.21* .14 .59**

Note. ICC � individual cognitive complexity; CC � cognitive complexity; N � 132 respondents.* p � .05. ** p � .01.


is controlled for. In line with this reasoning, werearranged the regression equation used to testHypothesis 3 as:

Y � b0 � b1(SDAICC) � b2(1

MeanAICC)

�b3CVAICC �c

where AICC is average cognitive complexitywithin groups. A similar strategy was used forthe time to reach consensus. The value for b3reflects the “clean” effect of cognitive disparityon the cognitive complexity of the groups. Theresults of the two rearranged regression equa-tions are presented in Table 6 (only the stan-dardized beta coefficients from Model 3 with allthe predictors included in the equation are pre-sented in the table). The results of the rear-ranged regression equations show support forHypothesis 3 in the sense that the effect ofcognitive disparity (corrected for the effect ofthe mean AICC) on GCC is positive and signif-icant. However, the results for the time to reachconsensus were not fully supported and the“clean” effect of disparity on the time groupsneed to reach agreement on the structure of thecognitive map is positive, but not statistically

significant. A possible explanation is the hiddeneffect of within unit moderating factors (e.g.,AICC interacts with cognitive disparity in thesense that the groups that need the most time toreach consensus are the groups with high AICCscores and high scores on cognitive disparity).

To further check the operationalization ofdisparity, we conducted an additional set ofanalyses. Starting from the alternative concep-tualization of group disparity, we heuristicallydivided our sample in three subgroups based onthe distribution of individual cognitive com-plexities within each group. In order to check ifthe relationship between cognitive disparity andgroup cognitive complexity is indeed linear, weplotted the scores of group cognitive complex-ity based on this distinction. The results arepresented in Figure 1.

The results using the second form of opera-tionalization show a reversed U-shaped relationbetween groups’ cognitive disparity andgroups’ cognitive complexity, with the highestlevel of complexity for groups with moderatelevels of cognitive disparity (see Figure 2).When the coefficient of variation was used asoperationalization of disparity, the relation be-tween the groups’ cognitive disparity andgroups’ cognitive complexity is negative andlinear. These differences raise the questionwhether the second operationalization of dispar-ity as presented in Harrison and Klein (2007) isaccurate.

Hypotheses 4 and 5 were the only hypothesesfor which the data were analyzed at the individ-ual level. In order to test these hypotheses, weartificially split the sample based on the differ-ence between the ICC and GCC. We used the

Table 4Results of the Regression Analysis for Time toReach Consensus

Model 1 Model 2

1. Gender variety .21 .202. Cognitive disparity .49***

R .21 .53**

Adj. R square .05 .25F change 1.95 13.49***

* p � .10. ** p � .05. *** p � .01

Table 5Results of the Regression Analysis for GroupCognitive Complexity

Group cognitivecomplexity

1. Gender variety .23*

2. Cognitive disparity �.26*

R .34*

Adj. R square .08

* p � .10. ** p � .05. *** p � .01

Table 6Results of the Rearranged Regression Equations forTime to Reach Consensus (TC) and GroupCognitive Complexity

TC (Model 3) GCC (Model 3)

1. SD AICC �.05 .792. 1/Mean AICC �.28 .103. CV AICC .60 �.92*

R .54*** .47***

Adj. R square .24 .16

Note. SD AICC � standard deviation average individualcognitive complexity; CV AICC � coefficient of variationfor AICC (cognitive disparity).* p � .10. ** p � .05. *** p � .01.


mean and standard deviation of the differencesbetween GCC and ICC to obtain three sub-groups: the first with an ICC lower than theGCC (lower than the mean minus a standarddeviation, N � 22), the second with an ICCapproximately equal to the GCC (one standarddeviation around the mean, N � 88), and thethird with an ICC higher than the GCC (higherthan the mean plus a standard deviation, N �22). Hypothesis 4 stated that group memberswith a higher CC than the group’s CC will beless satisfied about teamwork quality as com-pared with members with a lower individual CCthan the group’s CC. A t test for independentsamples was computed in order to test it. Groupmembers with a CC lower than the Group CCreport a higher satisfaction (M � 3.84, SD �.64) than the members with a CC higher thanthe group CC (M � 3.43, SD � 1.18). Thedifference, however, is not statistically signifi-cant t(42) � 1.42 ( p � .16); therefore, Hypoth-esis 4 is not supported. Hypothesis 5 stated thatthe participation to the task will be perceived asbeing less equal by group members with an ICChigher than the GCC as compared with mem-

bers with a lower ICC than the GCC. It wastested using a similar procedure as for satisfac-tion. The results support this hypothesis, show-ing that group members with a higher ICC thanthe GCC are less satisfied (M � 3.50,SD � 1.03) and perceive the individual contri-butions to the group task as less equal than thegroup members with an ICC lower than GCC(M � 4.04, SD � .53), t(42) � 2.19 ( p � .03).

In order to test Hypothesis 6, a hierarchicalregression was conducted. In the first step,group cognitive complexity was regressed toboth average individual cognitive complexityand quality of teamwork processes. The cross-product term was introduced in the second stepof the regression. Multicollinearity can be aproblem especially in small samples; therefore,in order to facilitate the interpretation of theresults and reduce the multicollinearity (see fordetails Aiken & West, 1991), the cross-productterm was computed based on the centered val-ues for teamwork quality and average ICC. Thecentered scores were computed by subtractingthe sample mean from the original values for

minimum medium maximum

Cognitive disparity

2,20

2,40

2,60

2,80

Me

an

sG

CC

Figure 1. Group cognitive complexity as a function of cognitive disparity operationalized asminimum, medium, and high.


teamwork quality and average ICC. The resultsare presented in Table 7.

The average individual cognitive complexityhas a positive effect on group cognitive com-plexity, a result which is consistent with theresults reported by Hendrick (1979). Further, asexpected, the quality of teamwork processes hasalso a positive impact on group cognitive com-plexity. The cross-product term introduced inthe second step also has a significant effect.Therefore, the impact of the interaction betweenaverage ICC and teamwork quality on GCC issignificant. As shown in Figure 2, the groupswith the highest cognitive complexity are thosewhose members have highly complex maps andexperienced effective teamwork processeswhile working as a group. In addition, we per-formed a t test for simple slopes to check for thedifference in the impact of average ICC depend-ing on the quality of teamwork. The t test forsimple slopes was computed by rearranging theregression equations into simple regressions ofgroup cognitive complexity on average ICC forconditional values of teamwork quality (see fordetails, Aiken & West, 1991). When the team-work quality is high (higher than the mean forthe entire sample), the impact of average ICCon GCC was positive and significant (b � .72,t � 4.55, p � .0001). When teamwork quality is

low (lower than the mean for the entire sample),the impact is still positive and significant, butmuch lower than in the first case (b � .45,t � 2.33, p � .03). Based on the results from thet test for simple slopes, it can be concluded thatthe positive impact of average ICC on GCC issignificantly accentuated under high teamworkquality conditions. Therefore, Hypothesis 6 isfully supported.

Discussion

In this study we set out to explore the impli-cations of group diversity and teamwork pro-cesses on the cognitive complexity of groups.More specifically, we examined to what extentgroup diversity (as variety and disparity), team-work processes, and the average ICC withingroups explain the complexity of the conceptualnetworks for collaboration developed bygroups. Concerning group diversity, we startedfrom a taxonomy introduced by Harrison andKlein (2007) and argued that group diversity asvariety (gender variety) is beneficial, whilegroup diversity as disparity is detrimental forGCC. We also argued that group disparity hasnegative implications for the individual groupmembers because it is associated with dissatis-faction and perception of unequal participationto the group task. With respect to GCC, weargued in line with other group cognition schol-ars that it is explained both by ICC and team-work processes. To examine these contentions,

1

1.5

2

2.5

3

3.5

4

4.5

5

Low average ICC High average ICC

MeansGCC

Low teamwork quality

High teamwork quality

Figure 2. The regression slopes for the interaction effectof teamwork processes quality and average individual cog-nitive complexity (CC) on group cognitive complexity.ICC � individual cognitive complexity; GCC � groupcognitive complexity.

Table 7Results of the Regression Analysis for theInteraction of Teamwork Processes and IndividualCognitive Complexity on Group CognitiveComplexity

Step and variable

Group CC

1 2

1. Quality of teamwork processes .43*** .45***

Average ICC .57*** .56***

2. Teamwork quality � Average ICC .23**

R .71*** .74***

Adj. R square .50 .54F change 20.91*** 3.74**

Note. ICC � individual cognitive complexity; CC � cog-nitive complexity, the cross-product term was computedbased on the centered values of quality of teamwork pro-cesses and average ICC.* p � .10. ** p � .05. *** p � .01.


we collected data from 44 groups using multiplemethods. Unlike most of the previous researchon group cognition, we evaluated GCC as agroup-level variable and used a cognitive map-ping technique to elicit and represent the con-ceptual networks for collaboration. Prior, thesame method was used to investigate the way inwhich individual group members make sense ofcollaboration. A questionnaire was used after-ward to investigate the quality of teamworkprocesses, group members’ satisfaction, andtheir perceptions of individual participation tothe group’s task.

As predicted by Hypothesis 1, we found thathighly diverse groups need more time to reachconsensus in a cognitive mapping session andthat the strongest impact is obtained for groupdiversity as disparity. When asked to create amap for collaboration, groups in which only oneindividual had a highly complex representationabout collaboration needed more time to reachconsensus than groups in which the cognitivecomplexity was evenly distributed. The asym-metrical distribution of a highly valued asset forgroups (individual cognitive complexity) leadsto long debates. This result is in line with thepredictions of conversion theory of minorityinfluence (Moscovici, 1980), which states thatthe presence of an informed (in the terms of ourarticle—highly complex) minority in a groupleads to a greater cognitive activity in the group,with the minority trying to persuade the othergroup members.

Groups high on disparity ended up withgroup cognitive maps that were less complexthan groups low on disparity. When looking atthe individual level, the group members with anICC higher than the GCC were less satisfied andperceived the contribution of the individualgroup members to the task as more unequal thandid those group members with an ICC lowerthan the GCC. At the systemic level, these re-sults suggest that group disparity is detrimentalto group effectiveness, both in terms of perfor-mance (GCC) and satisfaction. As theoreticallystated by Harrison and Klein (2007) group dis-parity is indeed associated with process losses(the correlation between the quality of team-work processes and cognitive disparity is�.32), with unequal participation to the task(r � �.28), and it creates frustration and dis-satisfaction among those individuals that have ahigh ICC. Although the grouping of individuals

according to their ICC relative to the GCCignores the impact of the within group interper-sonal interactions and interdependencies, theseresults can be used to argue that (1) the cogni-tive disparity was negatively perceived amonggroup members (see the lower scores on satis-faction and perception of equal participation)and (2) the conversion theory of minority influ-ence yields valid predictions concerning the im-plications of cognitive disparity on groups’ cog-nitive complexity. Therefore, future theoreticaldevelopments should connect these two previ-ously disparate research traditions: group diver-sity and minority influence, in order to furtherexplore the implications of cognitive disparityon groups’ cognitive complexity. Also, in amore practical vein, future research should ex-plore the ways in which the negative effects ofdisparity can be controlled for in real groups.Previous research shows that the negative ef-fects of expertise disparity on group perfor-mance in problem-solving tasks can be counter-balanced if the group members with a low ex-pertise will value the contribution of themember(s) with the highest expertise (see fordetails, Bonner, 2004). In conclusion, if cogni-tive disparity is acknowledged by the groupmembers and the most knowledgeable memberis allowed to take the lead, the negative impactof disparity can be attenuated.

Group variety has the opposite effects onGCC as compared to group disparity. In thisstudy gender diversity was used as a proxy forgroup variety in collaboration experiences andattitudes, with the assumption that men andwomen have qualitatively different experienceswith collaboration and so gender can be used asa variety-relevant attribute. Our results indeedshow that gender variety has a positive influ-ence on GCC. However, according to mostgroup diversity scholars (Milliken & Martins,1996; Pelled, 1996; Van Knippenberg et al.,2004; William & O’Reilly, 1998), gender isalso likely to trigger social categorization pro-cess and the use of stereotypes in interpersonalinteractions. Therefore, it is expected that gen-der diversity will ultimately impede group pro-cesses. The small and negative correlation be-tween gender variety and quality of teamworkprocesses offers partial support for this propo-sition. It can therefore be concluded that genderdiversity has a positive impact on GCC because(1) gender is in this instance an attribute highly


relevant for the task and (2) associated withvariety. Gender diversity, however, may havedisruptive effects for teamwork processes be-cause it triggers social categorization processesand therefore is also associated with separa-tion. One attribute can at the same time standmore than one type of diversity as describedby Harrison and Klein (2007). Therefore, themost important challenge for management ishow to highlight the positive effects ofknowledge variety and to reduce the negativeeffects of disparity or separation associatedwith a particular attribute (e.g., gender orrace).

The core result of this article refers to themoderator role of teamwork quality in the rela-tionship between average ICC and GCC. There-fore, it puts a key claim of group cognitionresearch to test. Previous research started fromthe assumption that the elaboration of task rel-evant information is increased if group mem-bers have complex knowledge structures (e.g.,Hendrick, 1979). Our study confirms these re-sults showing that the average ICC withingroups has a positive impact on GCC. However,the average ICC interacts with the quality ofteamwork processes in determining GCC, in thesense that groups with a high average ICC havethe highest GCC if they experience high qualityteamwork processes. The group members thathave complex representations about collabora-tion construct the most complex group concep-tual network for collaboration if they interact inan effective way and if they experience as fewprocess losses as possible.

Group cognition scholars argue that team-work processes are relevant for the developmentof team cognition (Rentsch & Woehr, 2004;Salas & Fiore, 2004). Group cognition emergesas a group-level phenomenon from the interplaybetween the knowledge structures of the groupmembers and the interaction processes that takeplace within groups (Curseu, 2006). Althoughgroup cognition has been theoretically de-scribed as a group-level phenomenon, it wasrarely investigated as such. Most of the researchfocused on aggregating group members’ cogni-tions in order to predict group performance andignored the interaction processes altogether.The present study went a step further and cap-tured group interactions both in the method toelicit and evaluate GCC and in the role of team-work processes in GCC. Therefore, GCC was

explored in the present study as a characteristicof group cognition that emerges from the inter-play between the individual knowledge struc-tures and the quality of interpersonal interac-tions inside the group.

Theoretical, Methodological, andPractical Implications

Based on our results it can be argued thatgroup disparity influences the group outcomesthrough teamwork processes, an argument thatis also in line with the open system modelsof group effectiveness (Gladstein, 1984;Hackman, 1987; McGrath, 1984). However, themediation models of group effectiveness are asimportant as the open system models becauseother factors (e.g., emergent states like cohe-sion, trust, potency) seem to mediate the impactof group diversity on group effectiveness (Ilgen,Hollenbeck, Johnson, & Jundt, 2005; Jung &Sosik, 1999; Marks, Mathieu, & Zaccaro,2001). Therefore, an interesting research direc-tion will be to test the extent to which teamworkprocesses and emergent states mediate the im-pact of group disparity on group effectivenessand how group disparity impacts on specificteamwork processes (see, e.g., the distinctionbetween transition phase processes and actionphase processes Marks et al., 2001), as well asthe emergent states.

As mentioned before, our study argues that adiversity attribute can be related to more thanonly one diversity type as described by Harrisonand Klein (2007). This raises the question ofhow to interpret an attribute exclusively interms of variety, disparity, or separation. Thepractical implication is that in order to increasethe GCC group leaders or managers should fo-cus on increasing variety as much as possibleand decreasing separation and disparity withintheir groups. Moreover, because a single at-tribute can be related to more than one type ofdiversity, it is important to note that group va-riety should not be increased in a way thatincreases disparity or separation simulta-neously.

Another issue concerns the alternative opera-tionalization of group disparity suggested byHarrison and Klein (2007). They suggest thatthe coefficient of variation is a good indicator ofdisparity because high values for the coefficient ofvariation are obtained if a single member in a


group has a high score on a certain scale, whileothers have low scores. The same arguments wereprovided by Bedeian and Mossholder (2000), whoshowed that the theoretical maximum for the co-efficient of variation is obtained when all cases ina group have zero values except one. On the otherhand, Harrison and Klein (2007) suggest anotheroperationalization for disparity (maximum dispar-ity if one group member is situated at the top ofthe scale, while the others are situated at the lowerlevels, medium disparity if all group members areequally distributed along the scale and minimumdisparity if the majority of the members are situ-ated at the top of the scale, or all members aresituated at the same level of the scale—low, me-dium or high). When we divided our sample ac-cording to this second operationalization of dis-parity, the pattern of results changed dramatically(see Figure 1). The results show that the relation-ship between disparity and GCC using this opera-tionalization of disparity has an inverted U shape.The small number of groups in the extreme groups(for low disparity, N � 4, and for high disparity,N � 13) suggests that these results should beinterpreted with caution. We conclude that theoperationalization of disparity is not yet com-pletely clear and future research is needed in orderto extensively test the two operationalizations inparallel.

Our study shows that GCC is a group-levelphenomenon that emerges from the interactionbetween the group members, and as a conse-quence, it depends both on the ICC of the groupmembers as well as on their interaction whileperforming the task. Any attempt to explore thecognitive complexity of groups should thereforesatisfy the criteria proposed by Bar-Tal (1990)for the evaluation of group-level phenomena.First, they should address the group as a whole.Second, group members’ agreement with regardto the construct must be demonstrated. Third,the construct must discriminate between groupsand finally the origin of the construct mustreflect group interaction processes (Bar-Tal,1990). The cognitive-mapping technique thatwas used to investigate GCC in the presentarticle satisfies all these essential criteria forgroup-level evaluations. In order to investigateGCC further, the use of this technique, as wellas other methods that satisfy the above men-tioned criteria, is welcomed.

The effect of the interaction between team-work processes and average ICC on GCC also

has straightforward practical implications. It isobvious that in order to get the best out of agroup it is necessary, but not sufficient, to bringtogether the most knowledgeable members. It isalso highly relevant to make sure that interac-tion processes that occur while working at thetask are highly effective. Process facilitationinterventions and measures might enhance theextent to which the group as a whole uses thecognitive resources of its members in order toget the best possible outcomes.

References

Aiken, L. S., & West, S. G. (1991). Multiple regres-sion: Testing and interpreting interactions. New-bury Park, CA: Sage.

Axelrod, R. (1976). The nature of decision. Prince-ton, NJ: Princeton University Press.

Bar-Tal, D. (1990). Group beliefs. A conception foranalyzing group structure, processes and behav-ior. New York: Springer-Verlag.

Bedeian, A. G., & Mossholder, K. W. (2000). On theuse of coefficient of variation as a measure ofdiversity. Organizational Research Methods, 3, 3,285–297.

Bieri, J. (1955). Cognitive complexity-simplicity andpredictive behavior. Journal of Abnormal and So-cial Psychology, 51, 263–268.

Bonner, B. L. (2004). Expertise in group problemsolving: Recognition, social combination, and per-formance. Group Dynamics, 8, 4, 277–290.

Bougon, M. (1983). Uncovering cognitive maps: The“Self Q” technique. In G. Morgan (Ed.), Beyondmethod: A study of organizational research strat-egies (pp. 173–188). New York: Sage.

Bower, C. A., Pharmer, J. A., & Salas, E. (2000).When member homogeneity is needed in workteams. A meta-analysis. Small Group Re-search, 31, 305–327.

Calori, R., Johnson, G., & Sarnin, P. (1994). CEO’scognitive maps and the scope of the organization.Strategic Management Journal, 15, 437–457.

Carley, K. M. (1993). Coding choices for textualanalysis: A comparison of content analysis andmap analysis. In P. V. Marsden (Ed.), Sociologicalmethodology (Vol. 23, pp. 75–126). Oxford, UK:Blackwell.

Carley, K. M. (1997). Extracting team mental modelsthrough textual analysis. Journal of Organiza-tional Behavior, 18, 533–558.

Cooke, N. J., Kiekel, P. A., Salas, E., Stout, R.,Bowers, C., & Cannon-Bowers, J. (2003). Measur-ing team knowledge: A window to the cognitiveunderpinnings of team performance. Group Dy-namics, 7, 179–199.


Cooke, N. J., Salas, E., Cannon-Bowers, J. A., &Stout, R. J. (2000). Measuring team knowledge.Human Factors, 42, 151–173.

Cooke, N. J., Salas, E., Kiekel, P. A., & Bell, B.(2004). Advances in measuring team cognition. InE. Salas & S. M. Fiore (Eds.), Team cognition.Understanding the factors that drive process andperformance (83–106). Washington, DC: Ameri-can Psychological Association.

Curseu, P. L. (2003). Formal group decision-making.A social cognitive approach. Cluj-Napoca, Roma-nia: ASCR Press.

Curseu, P. L. (2006). Emergent states in virtualteams. A complex adaptive systems perspective.Journal of Information Technology, 21, 249–261.

Curseu, P. L., & Rus, D. (2005). The cognitive com-plexity of groups. A critical look at team cognitionresearch. Cognitie, Creier, Comportament (Cogni-tion, Brain, Behavior) 9, 681–710.

Devine, D. J. (2002). A review and integration ofclassification systems relevant to teams in organi-zations. Group Dynamics, 6, 291–310.

Eby, L. T., Meade, A. W., Parisi, A. G., & Douthitt,S. S. (1999). The development of an individuallevel groupwork expectations measure and the ap-plication of a within-group agreement statistic toassess shared expectations for groupwork. Orga-nizational Research Methods, 2, 366–394.

Gladstein, D. L. (1984). Groups in context: A modelof task group effectiveness. Administrative ScienceQuarterly, 29, 499–517.

Gomez, A., Moreno, A., Pazos, J., & Sierra-Alonso,A. (2000). Knowledge maps: An essential tech-nique for conceptualisation. Data & KnowledgeEngineering, 33, 169–190.

Goodwin, V. L., Wofford, J. C., & Harrison, D.(1990). Measuring cognitive complexity in the or-ganizational domain. Paper presented at the annualmeeting of the Academy of Management. SanFrancisco, CA.

Hackman, J. R. (1987). The design of work teams. InJ. W. Lorsch (Ed.), Handbook of OrganizationalBehavior (pp. 315–342). Englewood Cliffs, NJ:Prentice Hall.

Harrison, D. A., & Klein, K. J (2007). What’s thedifference? Diversity constructs as separation, va-riety, or disparity in organizations. Academy ofManagement Review, 32(4).

Hendrick, H. W. (1979). Differences in group prob-lem solving behavior and effectiveness as a func-tion of abstractness. Journal of Applied Psychol-ogy, 64, 518–525.

Hinsz, V. B., Tindale, R. S., & Vollrath, D. A.(1997). The emerging conceptualization of groupsas information processors. Psychological Bulletin,121, 43–64.

Hoegl, M., & Gemuenden, G. H. (2001). Teamworkquality and the success of innovative projects: A

theoretical concept and empirical evidence. Orga-nization Science, 12, 435–449.

Hoffman, L. R., & Maier, N. R. F. (1961). Qualityand acceptance of problem solutions by membersof homogeneous and heterogeneous groups. Jour-nal of Abnormal and Social Psychology, 62, 401–407.

Hollan, J., Hutchins, E., & Kirsch, D. (2000). Dis-tributed cognition: Toward a new foundation forhuman-computer interaction research. ACMTransactions on Computer – Human Interac-tion, 7, 174–196.

Huff, A. S. (1990). Mapping strategic thought.Chichester, UK: Wiley.

Hutchins, E. (1995). Cognition in the wild. Cam-bridge, MA: MIT Press.

Ilgen, D. R., Hollenbeck, J. R., Johnson, M., & Jundt,D. (2005). Teams in organizations: From Input-Process-Output models to IMOI models. AnnualReview of Psychology, 56, 517–543.

James, L. R., Demaree, R. R. G., & Wolf, G. (1984).Estimating within-groups interrater reliability withand without response bias. Journal of Applied Psy-chology, 69, 85–98.

Jung, D. I., & Sosik, J. J. (1999). Effects of groupcharacteristics on work group performance: A lon-gitudinal investigation. Group Dynamics, 3, 279–290.

Kelly, G. A. (1955). The psychology of personalconstructs. New York: Norton.

Lagan-Fox, J., Code, S., & Langfield-Smith, K.(2000). Team mental models: Techniques, meth-ods, and analytic approaches. Human Factors, 42,242–271.

Marks, M. A., Mathieu, J. E., & Zaccaro, S. J. (2001).A temporally based framework and taxonomy ofteam processes. Academy of Management Re-view, 26, 356–376.

McGrath, J. E. (1984). Groups: Interaction and per-formance. Englewood Cliffs, NJ: Prentice Hall.

Milliken, F. J., & Martins, L. L. (1996). Searching forcommon threads: Understanding the multiple ef-fects of diversity in organizational groups. Acad-emy of Management Review, 21, 402–433.

Mohammed, S., Klimoski, R., & Rentsch, J. R.(2000). The measurement of team mental models:We have no shared schema. Organizational Re-search Methods, 3, 123–165.

Moscovici, S. (1980). Toward a theory of conversionbehavior. In L. Berkowitz (Ed.), Advances in ex-perimental social psychology (Vol. 13, pp. 209–239). New York, NY: Academic Press.

Okhuyzen, G. A., & Eisenhardt, K. M. (2002). Inte-grating knowledge in groups. Organization Sci-ence, 13, 370–386.

Pelled, L. H. (1996). Demographic diversity, conflict,and work group outcomes: An intervening processtheory. Organization Science, 7, 615–631.


Rentsch, J. R., & Woehr, D. J. (2004). Quantifyingcongruence in cognitions: Social relations model-ing and team member schema similarity. In E.Salas & S. M. Fiore (Eds.), Team cognition. Un-derstanding the factors that drive process and per-formance (pp. 11–31). Washington, DC: AmericanPsychological Association.

Rogelberg, S. G., & Rumery, S. M. (1996). Genderdiversity, team decision quality, time on task, andinterpersonal cohesion. Small Group Research, 27,79–90.

Sacco, J. M., & Schmitt, N. (2005). A dynamicmultilevel model of demographic diversity andmisfit effects. Journal of Applied Psychology, 89,1–29.

Salas, E., & Fiore, S. M. (2004). Why team cogni-tion? An overview. In E. Salas & S. M. Fiore(Eds.), Team cognition. Understanding the factorsthat drive process and performance (pp. 3–8).Washington, DC: American Psychological Associ-ation.

Schroder, H., Driver, M., & Streufert, S. (1967).Human information processing. New York: Holt,Rinehart, & Winston.

Schruijer, S. G. L., & Mostert, I. (1997). Creativityand sex composition: An experimental illustration.European Journal of Work and OrganizationalPsychology, 6, 175–182.

Schruijer, S. G. L., & Vansina, L. (1997). An intro-duction to group diversity. European Journal ofWork and Organizational Psychology, 6, 129–138.

Stuhlmacher, A. F., & Walters, A. E. (1999). Genderdifferences in negotiation outcome: A meta-analysis. Personnel Psychology, 52, 653–677.

Sundstrom, E., Busby, P. L., & Bobrow, W. S.(1997). Group processes and performance: Inter-

personal behaviors and decision quality in groupproblem solving by consensus. Group Dynam-ics, 1, 241–253.

Teachman, J. (1980). Analysis of population diver-sity. Sociological Methods and Research, 5, 341–362.

Van Knippenberg, D. van, De Dreu, C. K. W., &Homan, A. C. (2004). Work group diversity andgroup performance: An integrative model and re-search agenda. Journal of Applied Psychology, 89,1008–1022.

Weber, S. S., & Donahue, L. M. (2001). Impact ofhighly and less job-related diversity on work groupcohesion and performance: A meta-analysis. Jour-nal of Management, 27, 141–162.

Weick, K. E., & Roberts, K. (1993). Collective mindin organizations: Heedful interrelating on flightdecks. Administrative Science Quarterly, 38, 357–381.

Williams, H. M., & Mean, L. J. (2004). Measuringgender composition in work groups: A comparisonof existing methods. Organizational ResearchMethods, 7, 456–474.

Williams, K. Y., & O’Reilly, C. A. (1998). Demog-raphy and diversity in organizations: A reviewof 40 years of research. Research in Organiza-tional Behavior, 20, 77–140.

Wood, W., Lundgren, S., Ouellette, J. A., Busceme,S., & Blackstone, T. (1994). Minority influence: Ameta-analytic review of social influence processes.Psychological Bulletin, 115, 323–345.

Received June 8, 2006Revision received January 30, 2007

Accepted February 13, 2007 �


Appendix A Appendix B

A Taxonomy of Types of Relations Identifiedin the Cognitive Maps (adapted from Gomez,Moreno, Pazos, & Sierra-Alonso, 2000)

Gomez et al. (2000) discussed the process ofconceptualization as a critical element in everyproblem-solving activity. The authors presenteda framework for conceptual modeling in whichthey described several types of possible relationsbetween concepts (relations are interconnections be-tween concepts in a conceptual network). Seventypes of connections are particularly relevant for thecase of cognitive maps (for a detailed discussion, seeGomez et al., 2000, p. 174).

Causal relations (CA) describe how a given actionor phenomenon induces (determines) another state,action, or event (e.g., A is the cause of B, A needs B,A fires B, if A than B) or describe the conditions oractions are followed by consequences or reactions(e.g., A enables B, A needs B).

Association (ASO) describes how two or moreconcepts are correlated (e.g., A is related or associ-ated to B, A is connected to B, A is in contact withB) or describes combination of concepts (e.g., A andB are combined to. . .).

Equivalence (EQ) establishes the equality be-tween two or more apparently different concepts,including similarity (establishes which concepts aresimilar or analogue and to what extent; e.g., A �B � C, A is similar to B).

Topological (TOP) relations describe the spatialdistribution of concepts representing physical items(e.g., A is above B, A is to the right of B, A is insideB).

Structural (STR) relations describe how a conceptor a group of concepts can be decomposed into parts(also inclusion/exclusion relations, A is a part of B, Aand B are parts of C), or describe how several con-cepts share a common trait or are united by a com-mon element (A, B, and C share common elements).

Chronological (CHR) relations describe the waytwo or more concepts are related in a time sequence(e.g., A occurs before B, A and B occur simulta-neously, A occurs during B, A starts before B ends).

Hierarchical (HIE) relations describe the categor-ical relation between concepts (one or several ele-ments are subordinated to one or several others) (e.g.,A is subordinated to B; A, B and C are subordinatedto D), or describe taxonomic relations (e.g., A can beclassified as B, C, and D).

(Appendix continues)

Table 8A List of Concepts Related to CollaborationRanked on the Basis of Their Frequency in theFree Association Technique

Rank Concept Frequency

1. Teamwork 302. Understanding 223. Friendship 184. Communication 185. Support 166. Stress 147. Debates 138. Time spent together 119. Divergences 10

10. Tolerance 911. Satisfaction 912. Being informed (knowledgeable) 713. Frustration 714. Group 615. Unity 616. Influence 617. Cooperation 618. Meetings 619. Involvement 620. Interaction 621. Knowledge 522. Shared ideas 623. Effort 624. Seriousness 525. Achievement 526. Compromise 527. Patience 528. Roles 429. Cohesion 430. Rules 431. Trust 432. Consensus 433. Arguments 434. Innovation 435. Punctuality 436. Interdependence 337. Respect 338. Defensive reactions 339. Inflexibility 340. Conflict 3


Appendix C

RolesRulesInteractionInfluence

Group

Cooperation

Collaboration

MeetingsTeamworkSeriousnessInvolvementTime spent togetherPatienceDiscussions

KnowledgeBeing informed (knowledgeable)

Divergences

Punctuality

Success

Satisfaction

StressFrustrationDefensive reactionsInflexibility

Conflict

Arguments

CohesionInterdependencyRespectSupportToleranceTrustUnderstandingCompromisesCommunication

requireshas (develops)

is based on

means

equals

sometimesleadsto

Consensus

isassociated

with

leadstogenerates

through

leadto

generates

need

Includes/itisbasedon

Legend: The coding of the cognitive mapConcepts used =35Connections = 32Types of relations = 4 (ASO, CA, STR, EQ)

Complexity of the map = 65.335

4*32 =

Figure 3. Example of a group cognitive map, the coding scheme, and the formula tocompute cognitive complexity.


The effects of groups' variety and disparity on groups' cognitive complexity

Documents