20
Group Dynamics: Theory, Research, and Practice 2000, Vol. 4, No. 1,7-26 Copyright 2000 by the Educational Publishing Foundation 1089-2699/00/55.00 DOI: 10.1037//1089-2699.4.1.7 Group Cohesion: From "Field of Forces" to Multidimensional Construct Kenneth L. Dion University of Toronto Psychological research on cohesion stems from contributions in the 1940s and 1950s by Festinger and his colleagues, who defined cohesion as a "field of forces" acting on individuals to remain in the group. In the 1950s, critics of this definition noted that different cohesion measures often failed to intercorrelate. By the mid-1960s, A. J. Lott and B. E. Lott (1965) conceptualized cohesion as interpersonal attraction because researchers mainly focused on this "force." Multidimensional models of cohesion predominated in the 1980s and 1990s, the debate now focused on defining those dimensions. In addition to A. V. Carron's (1988) hierarchical model, several bidimen- sional models are summarized, including task and social cohesion, vertical and horizontal cohesion, belongingness and morale, and personal versus social attraction. Cohesion will continue to be a vital construct in research on groups and organizations into the 21st century, with important challenges to be addressed. The word cohesion derives from the Latin word cohaesus, meaning "to cleave or stick together." In physics and chemistry, cohesion refers to the force(s) binding molecules of a substance together. In psychology and the social sciences, a similar metaphor applies, with the term cohesion or cohesiveness describing the process(es) keeping members of a small group or larger social entity (e.g., military unit, business organization, ethnic group, or society) together and united to varying degrees. Other terms, such as attraction, morale, synlality, and solidarity, when applied to groups and organiza- tions of various sizes and their members, are also often used synonymously with cohesion (Dion, 1990). Similarly, because the construct of "affective tone" applies to groups as well as to individuals (George, 1990), a group's net affective tone provides an additional or more global affective element to the cohesion construct. An earlier version of this article was given as an invited presentation to the sixth SESP Groups preconference, Lexington, Kentucky, October 1998. Preparation of this article was facilitated by a research grant from the Social Sciences and Humanities Research Council of Canada as well as a sabbatical research leave from the University of Toronto. Correspondence may be directed to Kenneth L. Dion, Department of Psychology, 100 St. George Street, 4010 Sidney-Smith Hall, University of Toronto, Toronto, Ontario, Canada M5S 3G3. Electronic mail may be sent to DIONKLtg'psych.utoronto.ca. For the last 50 years, the cohesion construct has had its critics (e.g., Albert, 1953; Mudrack, 1989) and ups and downs within the two "social psychologies" of psychology and sociology, respectively (Stryker, 1977). In the 1960s, small-group researchers (e.g., Cartwright, 1968; Golembiewski, 1962; B. E. Lott, 1961) typically identified cohesion as one of the most important properties, if not the most important property, of groups. Since then, individual research studies, meta-analyses, and narrative literature reviews have shown that group cohesion correlates with other important group phenomena, such as small-group performance (e.g., Dion & Evans, 1992; C. R. Evans & Dion, 1991: Gully, Devine, & Whitney, 1995; Mullen & Copper, 1994), intergroup conflict (Dion, 1979; Stein, 1976), and therapeutic change (Yalom, 1995), as well as key indicators of organizational functioning, such as absenteeism (Keller, 1983), group performance in organizational contexts (Keller, 1986), and reported well-being (e.g., Bliese & Halverson, 1996). The cohesion construct is not the exclusive province of social psychologists interested in groups. Rather, the construct is shared with family, industrial-organizational, military, and sports psychologists (among others) within psychology, as well as with sociologists, anthro- pologists, and political scientists (among others) within the social sciences. All of these disci- plines and subdisciplines enrich and sustain

Group Cohesion: From Field of Forces to Multidimensional Construct

  • Upload
    others

  • View
    1

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Group Cohesion: From Field of Forces to Multidimensional Construct

Group Dynamics: Theory, Research, and Practice2000, Vol. 4, No. 1,7-26

Copyright 2000 by the Educational Publishing Foundation1089-2699/00/55.00 DOI: 10.1037//1089-2699.4.1.7

Group Cohesion: From "Field of Forces"to Multidimensional Construct

Kenneth L. DionUniversity of Toronto

Psychological research on cohesion stems from contributions in the 1940s and 1950s byFestinger and his colleagues, who defined cohesion as a "field of forces" acting onindividuals to remain in the group. In the 1950s, critics of this definition noted thatdifferent cohesion measures often failed to intercorrelate. By the mid-1960s, A. J. Lottand B. E. Lott (1965) conceptualized cohesion as interpersonal attraction becauseresearchers mainly focused on this "force." Multidimensional models of cohesionpredominated in the 1980s and 1990s, the debate now focused on defining thosedimensions. In addition to A. V. Carron's (1988) hierarchical model, several bidimen-sional models are summarized, including task and social cohesion, vertical andhorizontal cohesion, belongingness and morale, and personal versus social attraction.Cohesion will continue to be a vital construct in research on groups and organizationsinto the 21st century, with important challenges to be addressed.

The word cohesion derives from the Latinword cohaesus, meaning "to cleave or sticktogether." In physics and chemistry, cohesionrefers to the force(s) binding molecules of asubstance together. In psychology and the socialsciences, a similar metaphor applies, with theterm cohesion or cohesiveness describing theprocess(es) keeping members of a small groupor larger social entity (e.g., military unit,business organization, ethnic group, or society)together and united to varying degrees. Otherterms, such as attraction, morale, synlality, andsolidarity, when applied to groups and organiza-tions of various sizes and their members, arealso often used synonymously with cohesion(Dion, 1990). Similarly, because the constructof "affective tone" applies to groups as wellas to individuals (George, 1990), a group'snet affective tone provides an additional ormore global affective element to the cohesionconstruct.

An earlier version of this article was given as an invitedpresentation to the sixth SESP Groups preconference,Lexington, Kentucky, October 1998. Preparation of thisarticle was facilitated by a research grant from the SocialSciences and Humanities Research Council of Canada aswell as a sabbatical research leave from the University ofToronto.

Correspondence may be directed to Kenneth L. Dion,Department of Psychology, 100 St. George Street, 4010Sidney-Smith Hall, University of Toronto, Toronto, Ontario,Canada M5S 3G3. Electronic mail may be sent toDIONKLtg'psych.utoronto.ca.

For the last 50 years, the cohesion constructhas had its critics (e.g., Albert, 1953; Mudrack,1989) and ups and downs within the two "socialpsychologies" of psychology and sociology,respectively (Stryker, 1977). In the 1960s,small-group researchers (e.g., Cartwright, 1968;Golembiewski, 1962; B. E. Lott, 1961) typicallyidentified cohesion as one of the most importantproperties, if not the most important property, ofgroups. Since then, individual research studies,meta-analyses, and narrative literature reviewshave shown that group cohesion correlates withother important group phenomena, such assmall-group performance (e.g., Dion & Evans,1992; C. R. Evans & Dion, 1991: Gully, Devine,& Whitney, 1995; Mullen & Copper, 1994),intergroup conflict (Dion, 1979; Stein, 1976),and therapeutic change (Yalom, 1995), as wellas key indicators of organizational functioning,such as absenteeism (Keller, 1983), groupperformance in organizational contexts (Keller,1986), and reported well-being (e.g., Bliese &Halverson, 1996).

The cohesion construct is not the exclusiveprovince of social psychologists interested ingroups. Rather, the construct is shared withfamily, industrial-organizational, military, andsports psychologists (among others) withinpsychology, as well as with sociologists, anthro-pologists, and political scientists (among others)within the social sciences. All of these disci-plines and subdisciplines enrich and sustain

Page 2: Group Cohesion: From Field of Forces to Multidimensional Construct

DION

cohesion research as a continuing and thrivingenterprise, ensuring that the construct will continueto survive and thrive into the 21st century.

Social psychologists, however, bring some-thing distinctive to this intra- and interdisciplin-ary project: namely, a focus on the conceptuali-zation, measurement, and critical exploration ofthe validity of the cohesion construct, especiallyby laboratory experiments and controlled stud-ies attempting to limit, if not eliminate, extrane-ous variables. These issues are still critical tocohesion research today. If they are not properlyexplored and developed, cohesion research isseriously compromised, whether cohesion servesas an independent or a dependent variable inone's research, thinking, or both.

This article takes a selective, historical tour ofthe cohesion construct over the past century,documenting some highlights in a century-longquest to understand group cohesion. The cohe-sion construct has clear historical roots in boththe psychology and sociology of the later 19thand earlier 20lh centuries, as illustrated inwritings by Freud and Durkheim, respectively.Reflecting a lifelong interest in archaeology andanthropology, Freud considered in depth theissue of social cohesion, by which he meant theforces binding men together in groups (Wallace,1977). Freud's ideas reflected several recurrentthemes: the importance of the leader, the mythof the primal horde and father, and theimportance of aim-inhibited libidinal ties. Forexample, in his 1913 treatise Totem and Taboo,Freud described the "primal horde," a group ofsons ruled by a tyrannical father who jealouslyexcludes them from access to the females, andwent on to say: "One day the brothers who hadbeen driven out came together, killed anddevoured their father and so made an end of thepatriarchal horde . . . in the fact of devouringhim they accomplished their identification withhim, and each one acquired a portion of hisstrength" (pp. 141-142; cited in Wallace, 1977).Basically, Freud claimed the sons' attractiontoward the devoured father became transmutedinto a cohesion binding them together. Freud'sanalysis anticipated modern research on cohe-sion, such as the concept of vertical cohesion inmilitary psychology (described below), bysuggesting a close link between cohesion andleadership.

For his part, being concerned about increas-ing individualism in 19th-century France, soci-

ologist Emile Durkheim distinguished at amacrosociological level between mechanicaland organic solidarity (Driedger, 1996). Me-chanical solidarity characterized traditional ru-ral and folk societies, whose members weresimilar in background, values, and beliefs, thuscreating a strong, emotion-based sense ofcommunity. With mechanical solidarity, thegroup's norms are powerful forces constrainingits members. Organic solidarity, by contrast,develops from structural features of modernindustrial and bureaucratic society, such asdivision of labor and occupational specializa-tion. This structurally based, rational form ofsolidarity presumably results in a weaker formof collective conscience than mechanical solidar-ity, producing less uniformity and compliancefrom individuals in society as a result.Durkheim's (1897/1951) classic, sociologicalstudy of suicide rate variations among 19th-century European nations also linked suicide toanomie, a profound sense of social disintegra-tion arising from an absence of cohesion insocial structure (Hunt, 1985).

Modern psychological work on cohesion inthe 20th century, however, stems mainly fromthe research and theorizing of Kurt Lewin andhis students and colleagues (Marrow, 1969;Patnoe, 1988). In the late 1930s and early 1940s,Lewin laid the foundation for the concept ofgroup cohesion and the field-of-forces defini-tion—the first distinctly social psychologicalconceptualization of this construct. Lewin be-lieved that cohesiveness—the willingness toslick together—was an essential property ofgroups, without which they could not exist. Onthe basis of the belief that mathematical modelswould promote scientific psychology, Lewinespoused field theory and topological psychol-ogy, introducing concepts such as valence, forcefields, and life space into psychology. (Robert F.Bales [1999; Bales & Cohen, 1979] hascontinued to uphold the tradition of field theorywithin groups research to the present,)

Lewin defined cohesion as the set of forceskeeping members together, including both thepositive forces of attraction and the negativeforces of repulsion. He also introduced the termgroup dynamics into psychology (by which hemeant the positive and negative forces withingroups of people) and created the first graduateprogram devoted specifically to group dynamicsat the Massachusetts Institute of Technology

Page 3: Group Cohesion: From Field of Forces to Multidimensional Construct

SPECIAL ISSUE: GROUP COHESION

(MIT). Though not alone in emphasizing thepsychological importance of groups during the1940s, Lewin contended that the group(s) towhich one belongs are an important componentof individuals1 "life space," which clearlyanticipated social identity theory proposed threedecades later by Tajfel and Turner (1979).

As suggested above, Lewin's ideas influencedthe research and personnel at the Iowa ChildWelfare Research Unit in the 1930s and 1940sand the founding and establishment of MIT'sResearch Center for Group Dynamics from1945 until his death in 1947. The seminalstudies by Lewin, Lippitt, and White (1939) onthe impact of different leadership styles or"group atmospheres" included cohesion as oneof the first outcome measures. Lippitt's (1940)master's thesis research showed that greatergroup belongingness, as indexed in morefrequent use of the term we rather than /,occurred during discussions in democraticallyorganized boys' groups rather than autocrati-cally run ones. (In his lifelong study of Britishinternees in a German prison camp during WorldWar I, Ketchum [1965] likewise used thefrequency of we references in letters and campdocuments to reflect the rise and fall of thisprisoner society's solidarity and morale over theseveral years of the war's duration.) Later, atMIT's Research Center for Group Dynamics,Festinger's primary research interest in thepsychology of "group belongingness'1 reflectedLewin's influence and helped to launch field andexperimental research into group cohesivenessby social psychologists.

Cohesion as a Field of Forces

Perhaps Festinger's best-known cohesionresearch, in collaboration with then-graduatestudents Stanley Schachter and Kurt Back, wasthe famous study of two sets of housing units(Westgate and Westgate West) at MIT forreturning World War II veterans in the later1940s (Festinger, Schachter, & Back, 1950).Their nominal definition of cohesion was "thetotal field of forces which act on members toremain in the group" (p. 37). The strength of theforce field was, in turn, further defined by twoclasses of forces: (a) the attractiveness of thegroup and (b) the "means control" (i.e., theextent to which the group mediates goals for itsmembers)—anticipating a later distinction be-

tween social and task cohesion. A group'scohesion, in turn, defined and delimited itspower to influence members' behavior andattitudes, as well as to maintain and enforcegroup standards, if any.

Perhaps because the two housing complexeswere primarily social groups and Moreno (1934)had pioneered sociometric methods to assesspatterns of attraction in groups, Festinger et al.(1950) departed from their nominal definitionand instead operationalized cohesion in terms ofthe proportion of residents' choices from theirown housing court or building when listing thethree people in Westgate and Westgate West ofwhom they saw most socially. The higher theproportion of in-group choices from one'sparticular housing unit relative to total possiblechoices, the greater was the unit's cohesion. InWestgate and Westgate West, respectively, units*cohesion scores were correlated with the percent-age of "deviates," who were defined asresidents departing from their housing unit'sgenerally typical pattern of participation ornonparticipation in the tenants' organization,Westgate Council.

If the group used its power—that is, itscohesion—to influence its members to act andthink alike, then cohesion should relate posi-tively to the homogeneity of members' attitudesand behaviors, reflecting a group standard, andshould relate negatively to the percentage ofdeviates. Lack of a relation between cohesionand homogeneity presumably reflected theabsence of a group standard. The negative rankorder correlation of —.53 (p = .15, two-tailed)between housing unit cohesion and the percent-age of deviates in Westgate led Festinger et al.(1950) to infer the presence of a group standardin that housing complex: The higher the unitcohesion, the lower the percentage of deviates.For Westgate West, a smaller, nonsignificantcorrelation of - .20 indicated the absence of agroup standard.

Festinger et al. (1950) also carefully distin-guished group cohesion within the housing unitsfrom subgroups in the form of friendship cliquesmaking reciprocal choices within a particularhousing unit. Though these friendship sub-groups might be highly cohesive, the housingunit of which they were part would presumablynot be as cohesive with such cliques withinthem. Subtracting "in court" choices by half thenumber of mutually reciprocated choices served

Page 4: Group Cohesion: From Field of Forces to Multidimensional Construct

10 DION

to correct for the "subgroup" problem. With thiscorrection factor, the negative rank ordercorrelation between cohesion and percentagedeviates increased to — .74 (p — .02) in West-gate, a significant and stronger relationship thanwithout the correction, but it remained unchangedand nonsignificant at - .27 in Westgate West.

The MIT housing study set an agenda forexperimental research on cohesion in thecontext of social communication theory byFestinger (1950) and his colleagues in the late1940s and early 1950s. Schachter's (1951)"deviates" study, which showed greater commu-nication toward opinion deviates in highlycohesive groups during early phases of groupinteraction, followed by their later rejection inthe absence of conformity, is a well-knownexample. Basically, in social communicationtheory, high cohesion was hypothesized toinstigate greater communication among groupmembers, creating greater uniformity of opinionand behavior between and among them. Tocreate experimental variations in group cohesive-ness, Back (1951) varied information concern-ing the alleged likability or congeniality of othergroup members (i.e., interpersonal attraction),the desirability or interest value of the grouptask (i.e., task attraction), or the prestige of thegroup. Back compared these three types ofcohesion inductions and found no differences insocial influence toward a partner as a function ofthe type of cohesion. Cohesiveness, howeverinduced, was subsequently portrayed as unitaryin nature, and different cohesiveness inductionswere assumed to be "equivalent in [their]effects" by proponents of social communicationtheory (e.g., Schachter, 1951).

The original field-of-forces definition ofcohesion also shifted, subtly but importantly,from the total field of forces to the resultant offorces acting on members to remain in a group.This change in conceptual definition reflected ashift in focus from cohesion's causes to itseffects. Assessed at the individual level, theresultant forces acting on each member toremain in the group were subsequently termedattraction to group (Libo, 1953). N. J. Evansand Jarvis's (1986) 20-item questionnaire mea-sure of attraction to group assesses "an individu-al's desire to identify with and be an acceptedmember of the group" (p. 204).

The Gross Cohesion Questionnaire(GCQ)—an instrument sometimes described as

the most widely used cohesion measure in theliterature (Stokes, 1983) and still the object ofsome research attention (Cota, Dion, & Evans,1993, 1995)—was devised by then-Harvardstudent E. F. Gross (1957) on the basis of afield-of-forces definition of cohesion. Stokes(1983, p. 167, Table 1) listed seven GCQ itemsand two additional items developed by Yalornand Rand (1966). Sample GCQ questions formeasuring cohesiveness are: "How well do youlike the group you are in?" and "How attractivedo you find the activities in which youparticipate as a member of your group?"

N. Gross and Martin (1952) criticized thefield and experimental studies of cohesionconducted by Festinger and his colleagues,especially the former. They emphasized thewide gulf between the nominal definition (i.e.,the total field of forces acting on individuals toremain in the group) and the actual, operationaldefinition (the proportion of in-group choicesfor socializing) of cohesion in the MIT housingstudy, as well as the difficulty, if not impossibil-ity, of ever fully operationalizing this nominaldefinition by assessing all of the forces acting onmembers to stay in their groups.

N. Gross and Martin (1952) also related theWestgate study's operational definition to twoother cohesion indices—the ratio of dislikedgroup members and of isolates, respectively—in13 women's university residence houses from astudy of their own. Because the dislike andisolate ratios each reflect low cohesion, onewould expect them to correlate negatively withthe proportion of in-group choices but positivelywith one another. However, Gross and Martinnoted differences in a given group's rankingsacross these putative indices of group cohesion.The rank-difference correlations were - .37between the proportion of in-group choices andthe dislike ratio, —.42 between the proportion ofin-group choices and the isolate ratio, and .69between the dislike and isolate ratios, with onlythe last attaining statistical significance.

Although the pattern of correlations followedexpectations for the hypothesis of a unitarydimension of cohesiveness, that there was onlyone significant correlation was taken as indicat-ing deficiencies in the cohesion construct. N.Gross and Martin (1952), however, failed tonote that their limited sample of groups meantthat their test of interrelationships amongdifferent cohesion indices had low statistical

Page 5: Group Cohesion: From Field of Forces to Multidimensional Construct

SPECIAL ISSUE: GROUP COHESION 11

power. Even assuming a "large" effect size(viz., an r of .50 between a pair of cohesionmeasures), these investigators would have neededat least 28 groups—more than twice the sampleof groups they used—to have a power of .80, astandard recommended level today, and atwo-tailed alpha of .05 (Cohen, 1992).

N. Gross and Martin (1952) suggested twoalternatives for assessing group cohesion. Theyproposed, plausibly enough, directly askingmembers how attractive the group was on arating scale whose endpoints ranged from slightor no attractiveness to very great attractiveness.They also suggested indexing cohesion by agroup's resistance to disruptive forces. Nearly40 years later, Brawley, Carron, and Widmeyer(1988) operationalized this latter definition ofcohesion by having sport team members re-spond to hypothetical disruptive incidents. Theyexplored the relationship between cohesion andgroup resistance to disruption in two separatestudies of athletes and sports groups and found thatgroup task cohesion related positively to perceivedgroup resistance to disruption in both studies.

Eisman (1959) also tested N. Gross andMartin's (1952) hypothesis of little or no allegedrelationship among operational definitions ofcohesiveness by obtaining five putative mea-sures of cohesion from 14 "natural" studentgroups at the University of Colorado: 10fraternities and sororities, 1 academic club, and3 religious clubs. The five measures included (a)a sociometric index of friendship (specifically,the number of in-group choices when respon-dents listed their best friends on campus),similar to that typically used in previouscohesion research; (b) the "direct" rating ofgroup attractiveness Gross and Martin hadsuggested as a cohesion index; (c) the meannumber of 21 items checked as reasons forbelonging to the group; (d) the number of these21 items checked by more than half therespondents in a given group; and (e) valueconcordance among group members, as as-sessed by a revised version of Allport, Vernon,and Lindzey's (1951) value scales.

Eisman (1959) emphasized that the rankcorrelations (Kendall's tau) computed amongthese five cohesion indices failed to yield even asingle instance of a significant correlation,though several approached significance. Ofcourse, with only 14 groups, her study possessedlow statistical power (i.e., the probability of

obtaining a statistically significant result andrejecting a null hypothesis when it is invalid). Aswith N. Gross and Martin's (1952) study, evenassuming a large effect size for intercorrelationsof the cohesion measures, Eisman would haverequired at least twice her sample of groups in orderto have an adequate level of statistical power.

The Gross-Martin cohesion question in Eis-man's (1959) study fared poorly as well,correlating negatively with the sociometricindex and measures of reasons for belonging tothe group. Likewise, the value-concordancemeasure correlated negatively with all othermeasures, save for sociometric choice, but thismeasure is perhaps questionable in retrospect asan indicator of cohesion. However, the correla-tion between the sociometric index and thenumber-of-reasons measure came close to beingsignificant and probably would have been so,had more groups been sampled. In any case,Eisman (1959) interpreted her results as support-ing N. Gross and Martin's (1952) concerns andpointed out the difficulties of forming generaliza-tions about cohesion when different measuresare used across investigations. Perhaps mostimportant, she concluded that cohesion mightnot be a unitary or unidimensional concept afterall and suggested that Festinger's nominaldefinition of cohesion needed re-examination.

Noting that most researchers of cohesionfocused only on the one "force" of interpersonalattraction, Albert J. Lott and Bernice Lott (neeEisman; 1965) redefined cohesion as a groupproperty to be inferred from the number andstrength of mutually positive attitudes amonggroup members. Because little was known aboutthe antecedents of cohesion but a good deal wasknown about antecedents of liking, conceptual-izing cohesion as interpersonal attraction ap-peared to offer considerable conceptual lever-age. The conceptualization of group cohesion asinterpersonal attraction has since been criticizedby advocates of self-categorization theory fordefining a group-level concept in terms of aninterpersonal one and allegedly confusing thetwo levels of analysis. However, in their articleLott and Lott (p. 302) carefully noted:

We do not intend to throw out group concepts and tosubstitute for them concepts definable in terms of thebehavior of individuals, but we believe it to be bothmore parsimonious and systematic to begin with thelatter, our only source of data, after all.

Page 6: Group Cohesion: From Field of Forces to Multidimensional Construct

12 DION

Multidimensional Conceptualizationsof Cohesion

An alternative tradition of denning cohesionas a multidimensional construct arose initiallyfrom factor analytic studies of group propertiesand cohesion scales as early as the 1940s(Cattell, 1948; see Cota, Longman, Evans, Dion,& Kilik \ 19951 concerning the uses and misusesof using factor analysis to define and conceptual-ize cohesion) and subsequently by theoreticalanalysis. With a sample of 21 groups, Cattell andWispe (1948) found five factors, three of which(including the largest factor) reflected cohesionand morale. With 93 personality and groupperformance variables from a sample of 80groups (60 military and 20 student), Cattell,Saunders, and Stice (1953) extracted 15 factors,of which 4 reflected aspects of cohesion.

Measuring 61 group-related properties amongmembers of women's university residencegroups, Selvin and Hagstrom's (1963) centroidfactor analysis yielded five interpretable factors,the first two being distinctive features ofcohesion that they labeled Social Satisfactionand Sociometric Cohesion. High scores onSocial Satisfaction described groups whosemembers were satisfied with their campus life'ssocial aspects, including their own residence,and this factor resembled the "attraction to thegroup" concept in group dynamics research ofthe time. The second factor, Sociometric Cohe-sion, was denned by items reflecting the extentto which members' friendships and interactionswere centered in the group and their length ofgroup membership.

In military psychology, cohesion is a topic ofmajor interest (see Siebold, 1999), and militaryunit cohesion is a highly prized commodity.Accordingly, military psychologists have con-ducted a number of factor analytic studies ofcohesion and developed several multidimen-sional models of cohesion. Griffith (1988) wasinterested in clarifying the conceptual andoperational definitions of military unit cohesionwithin U.S. Army units and factor analyzeditems from questionnaire measures of cohesionand other group dimensions. Four factorsresulted but were further simplified as reflectingtwo "fundamental" dimensions at a theoreticallevel: (a) the direction of cohesion, contrastingvertical cohesion (superior-subordinate rela-tions) and horizontal cohesion and (b) the

Junctions of cohesion, contrasting instrumentalor task cohesion relating to task performanceand affective cohesion relating to interpersonaland emotional support.

Vertical and Horizontal Cohesion

Bliese and Halverson (1996) focused specifi-cally on vertical and horizontal cohesion as tworelated constructs reflecting relationship qualitywithin military units. Vertical cohesion refers tosubordinates' perceptions of their leaders' com-petence and considerateness. (As noted above,vertical cohesion reflects the link betweencohesion and leadership as well as the impor-tance of competent leadership in fostering groupcohesion.) The 11-item vertical cohesion scaleused in their analyses had a coefficient alpha of.89 and items such as "My officers are interestedin my personal welfare," and "The [noncommis-sioned officers] in this company would lead wellin combat." The 8-item horizontal cohesionscale had a coefficient alpha of .83, and sampleitems such as "I spend time when not on dutywith people in this company," and "My closestrelationships are with the people 1 work with."Basically, horizontal cohesion is similar tosocial or interpersonal cohesion. Confirmatoryfactor analyses (CFAs) showed that the verticaland horizontal cohesion scales are distinct butpositively correlated with one another, sharingapproximately 20% of their variance.

With the individual as unit of analysis, bothvertical and horizontal cohesion correlatedpositively (rs of .43 and .24, respectively) with ameasure of well-being in a sample of 99 U.S.Army companies. Bliese and Halverson (1996)also determined the proportion of the cohesion-well-being relationship that was due to indi-vidual and group effects, respectively. Forwell-being and vertical cohesion, 88% of therelationship was due to individual effects, and12% was due to group effects. For well-beingand horizontal cohesion, 94% of the relationshipwas due to the individual component, and 6%was due to the group component. Thus, becausemost of the relationship between cohesion andwell-being was due to individual effects, Blieseand Halverson were warranted in using individu-als as the unit of analysis in their research.

Page 7: Group Cohesion: From Field of Forces to Multidimensional Construct

SPECIAL ISSUE: GROUP COHESION

Carron 's Hierarchical Model

Sports psychology is another domain inwhich cohesion is unquestionably important. Atheory-driven approach to cohesion in sportsand exercise groups was begun in the earty-mid-1980s, and has continued since then, by sportspsychologist Albert Carron, in collaborationwith his associates. Carron (1988; Carron,Widmeyer, & Brawley, 1985, p. 248) and hisassociates proposed a hierarchical model ofcohesion that is graphically portrayed in Figure1. Group cohesion is a general construct at thepeak of the model, which is first differentiatedinto an individual component (individual attrac-tion to group) and a group component (groupintegration) at the middle level; in turn, eachmiddle-level component is further differentiatedinto task versus social subcomponents at thethird level at the model's base, resulting in fourfactors at the base that are assumed to bepositively correlated with one another.

For Carron and his colleagues, "group" and"individual" components refer to two objects orfoci of perception, as seen from the perspectiveof individual group members. The group-

individual distinction has often been suggestedin the cohesion literature (e.g., Cattell, 1948;Van Bergen & Koekebakker, 1959). Likewise,the distinction between task versus socialcohesion has clear precedence within groupsresearch. Bales's (1950, 1999; Bales & Cohen,1979) theories and methods for studying groupshave emphasized differentiating task-related andsocioemotional behaviors of group membersduring interaction. Similarly, several othergroup researchers independently urged theseparate measurement of the task and socialcomponents of cohesion (Mikalachki, 1969;Tziner, 1982a, 1982b; Wheeless, Wheeless, &Dickson-Markman, 1982).

The group-individual and task-social distinc-tions together yield the four subtypes ofcohesion of the hierarchical model (see Figure1); Group Integration-Social (Gl-S), GroupIntegTation-Task (GI-T), Individual Attractionsto Group-Social (ATG-S), and Individual Attrac-tions to Group-Task (ATG-T). Carron andcolleagues developed an 18-item Group Environ-ment Questionnaire (GEQ) that they have sinceused to assess perceived cohesion by individual

GROUP COHESION

INDIVIDUALATTRACTION

TO GROUP

GROUPINTEGRATION

Figure 1. Carron's (1988; Carron, Widmeyer. & Brawtey, 1985) hierarchical model of groupcohesion.

Page 8: Group Cohesion: From Field of Forces to Multidimensional Construct

14 DION

members and the four components of thehierarchical model (for descriptions of theGEQ's development, see Carron, Widmeyer, &Brawley, 1985).

The "individual" ATG items reflect members'perceived personal involvement with the group'stask and social dimensions. An example ofATG-S is "For me this team is one of the mostimportant social groups to which I belong." AnATG-T example is "I'm not happy with theplaying time I get" (reverse scored). GI itemsassess members' perceptions of the "similarity,closeness, and bonding within the team as awhole" in regard to its task and social aspects. Asample Gl-S item is "Our team would like tospend time together in the off season." Anexample of a GI-T item is "Our team is united intrying to reach its goals for performance."

Internal consistency reliabilities (i.e., coeffi-cient alpha) for GEQ scales from several studiesin the 1980s and 1990s are shown in Table I.With Canadian sports teams, subscale reliabili-ties have generally been adequate, though thishas not always been so with studies of sportsteam athletes from the United States, NewZealand, or Australia. Clearly, the number ofitems in the GEQ's four subscales is limited, soit is always wise to check their reliabilities.Although GEQ item content is framed for sportsteams, one can rephrase items to fit some othergroup types. The best subscale reliabilities in Table1 come from a GEQ scale modified for universityresidence groups (Carron & Ramsay, 1994).

Interscale correlations for the GEQ should bepositive but moderate. Indeed, they vary from

Table 1Internal Consistency Reliabilities for GroupEnvironment Scale Subscales in SelectedStudies (Numbered 1-5)

Scale

ATG-TATG-SGITGl-S

No. items

4554

1

.75

.64

.70

.76

2

.81

.75

.73

.83

3

.74

.40

.60

.67

4

.55

.21

.74

.67

5

.65

.60

.64

.72

Note. 1 = Carron, Widmeyer, and Brawley (1985);participants: athletes from Canadian college sports teams;2 = Carron and Ramsay (1994); participants: Canadianuniversity residence group members; 3 = Prapavessis andCarron (1996); participants: New Zealand sports teams; 4 =Prapavessis and Carron (1997b); participants: New Zealandsports teams; 5 = Schutz, Eom, Smoll, and Smith (1994);participants: U.S. high school athletes; ATG = attraction togroup; T — task; S = social; GI ~ group integration.

.28 to .41 for the sample of 247 athletes from 26diverse sport type teams used to define the final18-item version of the GEQ (Carron, Brawley,& Widmeyer, 1985). The highest intercorrela-tions are between the two task scales (.41) andthe two social scales (.40). By contrast, the twoATG scales' correlation is .28 and that for thetwo GI scales is .30. Though Carron and hiscolleagues do not do so themselves, one couldpool the task and social scales, respectively, andcollapse across the individual and group scalesto create a more reliable measure defined bymore items for task and social cohesion.

Considerable evidence of the GEQ subscales'validity with various criteria exists for sportsteams, exercise classes, and physical activitygroups. These criteria include perceived resis-tance to disruption of the group (Brawleyet al., 1988), individual work output (Prapaves-sis & Carron, 1997a), perceptions of groupgoal-related variables (Brawley, Carron, & Wid-meyer, 1993), competitive state anxiety(Prapavessis & Carron, 1996), self-responsibil-ity attributions for group success and failure,participation in team versus individual sports,tenure length of athletes in individual sportteams (Brawley, Carron, & Widmeyer, 1987),group size (Carron & Spink, 1995), sacrifice andconformity to norms (Prapavessis & Carron,1997b), and adherence to exercise regimens(Carron, Hausenblas, & Mack, 1996; Carron,Widmeyer, & Brawley, 1988; Spink & Carron,1994). The GEQ also shows evidence ofconvergent and divergent validity (Brawley etal., 1987).

The stability of the GEQ's factor structureand its factorial validity, however, remain to bedemonstrated for sports teams and other types ofgroups. Some evidence suggests that the GEQ'sfactor structure varies for groups other thansports groups and even for sports teams otherthan collegiate ones. Dyce and Cornell (1996)administered the GEQ to 315 members (most ofthem male) of 84 Canadian musical groups,when they were together as a group. Theiroblique factor analysis, using an eigenvalue-greater-than-1 rule for retaining factors, pro-duced three factors. The first factor consisted ofeight sociability items, the second factor com-prised seven dissatisfaction items, and the thirdfactor included three performance items withonly marginal internal consistency. Dyce andCornell also conducted oblique factor analyses

Page 9: Group Cohesion: From Field of Forces to Multidimensional Construct

SPECIAL ISSUE: GROUP COHESION 15

that prespecified four factors for the solution.With four factors prespecified, the solution wasreportedly difficult to interpret and did not resultin the theoretical structure proposed by Carronand his associates. On the other hand, presepeci-fication of a two-factor structure was interpret-able, with the two factors being Social Cohesion(8 items) and Task Cohesion (10 items), insupport of the task-social cohesion distinction.The group-individual distinction, by contrast, wasnot supported in Dyce and Cornell's analyses.

Perhaps Dyce and Cornell's (1996) findingssimply indicate the GEQ's inapplicability to atype of group (viz., musicians' bands) other thanthat originally intended by its creators (viz.,athletes in sports teams). However, with asample of U.S. high school athletes of bothsexes from team and individual sports, Schutz,Eom, Smoll, and Smith (1994) also questionedthe GEQ's factorial validity, using both explor-atory and confirmatory factor analytic tech-niques. One exploratory procedure was aprincipal-components analysis with obliquerotation and four factors prespecified, which ac-counted for only 31 % of the variance and yielded apattern of item loadings diverging in severalrespects from the GEQ's four-factor model.

Because an explicit theoretical model under-lies the GEQ, CFA allows researchers to assessthe GEQ's factorial and construct validity moredefinitively and directly than exploratory factoranalyses. However, two recently publishedreports of CFAs on GEQ data from athletic teamsamples disagree with one another. Schutz et al.(1994) conducted several CFAs on GEQ datafrom their high school athlete sample, includinga test of the first-order four-factor model; multi-group analyses as a function of gender and teamversus individual sports; and alternative models,including higher order ones. Using various goodness-of-fit indices, they concluded that none of themodels tested successfully fit the data.

By contrast, with a U.S. sample of intercolle-giate athletes from team sports, Li and Harmer(1996) reported CFAs supporting the GEQ'stheoretical model in several regards. For ex-ample, at the first-order level (i.e., the base ofthe model depicted in Figure 1), the first-orderfour-factor measurement model defined by theGEQ's theoretical model provided a demonstra-bly better fit than alternative one- and two-factormodels. Moreover, higher order CFAs providedevidence that the second-order "group" and

"individual" components of cohesion are latentvariables determining the four first-order factorsand providing better fit than some alternativemodels.

Li and Harmer (1996) concluded that theGEQ has factorial validity as a measure ofcohesion for intercollegiate athletes in teamsports and suggested that Schutz et al.'s (1994)results might be due to the fact that the sampleconsisted of high school athletes rather than thecollegiate ones from which the GEQ wasoriginally conceptualized and developed. How-ever, numerous other differences between thesestudies, singly or in combination, could easilyhave contributed to their divergent findings andconclusions concerning the GEQ's factorialvalidity as a cohesion measure. In addition to theeducational level and age of the athletes, thesedifferences include the size of the sample, theinclusion of individual sports as well as teamones, when the GEQ was administered relativeto the team's season of play, different standardsfor defining adequate fit for the CFA models,and whether the GEQ's intercorrelations ofordinal scale scores were first transformed topolychoric correlations prior to performing theCFAs, among other things.

Thus, a firm conclusion concerning theGEQ's factorial validity and the adequacy of itsunderlying theoretical model (viz., the hierarchi-cal model) await more research and consistentfindings, as well as additional studies, to definethe type of group and other boundary conditionsfor the GEQ as a cohesion measure for sportsand other groups. Unfortunately, the full hierar-chical model proposed by Carron and hiscolleagues cannot be tested beyond the secondorder using higher order CFAs. A single, higherorder factor requires at least three subfactors(and preferably more) at the immediatelypreceding, lower level in order for the measure-ment model to be sufficiently identified so as topermit a higher order CFA (see Kline, 1998, pp.233-236; Rindskopf & Rose, 1988). Becausethe latent variable of "group cohesion" at thethird order (i.e., the peak) of the hierarchicalmodel (see Figure 1) is conceptualized ascomprising only two factors at the second order(viz., group integration and individual attractionto group), the upper portion of the model cannotbe assessed by a higher order CFA.

Page 10: Group Cohesion: From Field of Forces to Multidimensional Construct

16 DION

Task Versus Social Cohesion

Independently of research by Carron and hisassociates in sports psychology, Stephen Zac-caro and his colleagues have themselves pro-posed a bidimensional conceptualization ofcohesion that differentiates task and interper-sonal cohesion. In several studies, they haveshown that differentiating these two types ofcohesion improves the prediction of group taskperformance. Zaccaro and Lowe (1988) foundthat task cohesion more strongly facilitatedgroup performance than did interpersonal cohe-sion on an "additive" task (see Steiner, 1972,for a typology of group tasks) that pooledindividuals' independent performances. In an-other study with a "disjunctive" task requiringgroup interaction (viz., a survival task in whichgroup members ranked items important to groupsurvival), the highest group performance wasobtained only when high levels of both task andsocial cohesion existed (Zaccaro & McCoy, 1988).

In a study of a cadet corps at a large,southeastern U.S. university, Zaccaro (1991)also demonstrated the "nonequivalent effects"of these two dimensions of cohesion with regardto four outcomes: group performance processes,role uncertainty, absenteeism, and individualperformance. In other words, he sought tosupport his multidimensional perspective oncohesion by showing differential patterns ofrelationship to this criterion set. He predictedthat task cohesion would correlate more highlywith these criteria, on the basis of the premisethat task cohesion heightens conformity toattendance norms and clarifies behavioral normsconcerning role performance. Task cohesionwas assessed with a 7-point scale on whichrespondents indicated their agreement withstatements that (a) they liked belonging to theirgroup because of the activities in which theyparticipated and (b) they did not like what theydid as group members (reverse scored; a — .62).Interpersonal cohesion was indexed by itemsstating that (a) they generally did not get alongwith their fellow group members (reversescored) and (b) they enjoyed belonging to theirgroup because they were friends with many oftheir group members (a — .65). Adjusting forgroup-level effects, Zaccaro found higher corre-lations between task cohesion and all criteriasave group performance than for interpersonalcohesion. This stronger relationship for task

cohesion stayed intact even when interpersonalcohesion was partialed out.

Reflecting yet another example of its useful-ness, Bernthal and Insko (1993) applied thesocial-task cohesion distinction to the group-think phenomenon. They proposed that Janis's(1982) groupthink model applies to groups andcontexts in which socioemotional cohesion isespecially prominent and that cohesion isdenned as attraction to other group membersrather than in terms of the task. To test thiscontention, they experimentally and orthogo-nally manipulated high and low levels of bothtask and socioemotional cohesion, respectively,in groups of undergraduate women given aconcept-formation task to perform. On postinter-action ratings reflecting groupthink symptoms,they supported their hypotheses that groupthinktendencies were least apparent when socioemo-tional cohesion was low but task-oriented cohesionwas high. They also noted that many of thetechniques Janis recommended to counteract group-think are, in effect, tactics that promote taskcohesion in groups. So, it is possible for groups tohave high cohesion, at least of the task variety,without promoting groupthink tendencies.

The roles of socioemotional cohesion inpromoting groupthink and of task cohesion ininhibiting it have also been explored by Mullen,Anthony, Salas, and Driskell's (1994) ineta-analysis of nine studies available prior to August1992 (thus, not including Bernthal & Insko,1993) that investigated the effects of cohesionon group decision-making quality. Cohesionmanipulations were rated for the extent to whichthey were based on interpersonal attraction,commitment to task, and group pride, respec-tively. Using the aforementioned sample ofstudies, Mullen et al. found no significantrelationships between group cohesion and deci-sion quality, whether cohesion was due tointerpersonal attraction, task commitment, orgroup pride. However, when they focused on asmaller sample of eight hypothesis tests inwhich other antecedents of groupthink were"neither exaggerated nor minimized," the differ-ential effects of socioemotional and task cohe-sion on quality of group decision became moreapparent. Specifically, where cohesion primarilyreflected interpersonal attraction, increasingcohesion was associated with lower decisionquality. By contrast, with cohesion inductionsreflecting task commitment or group pride,

Page 11: Group Cohesion: From Field of Forces to Multidimensional Construct

SPECIAL ISSUE: GROUP COHESION 17

greater cohesion was associated with improveddecision making.

Perceived Cohesion: Belongingnessand Morale

Another two-factor or bidimensional concep-tualization of cohesion, proposed by Bollen andHoyle (1990), links both of the "psychological"and "sociological" social psychology ap-proaches and extends the applicability of thecohesion construct beyond the confines of thelaboratory to larger groups, such as communi-ties. Noting that there is at present no generallyaccepted theoretical or operational definition ofthe cohesion construct, Bollen and Hoyledifferentiated "objective" and "subjective'1

approaches to the cohesion construct. Theobjective approach consists of taking a compos-ite of each group member's judgments orfeelings toward every other group member, withjudgments in the form of, say, sociometricchoices. The "subjective approach," by con-trast, probes individual group members' percep-tions of a specified group's cohesion. Exemplify-ing a subjective approach, perceived cohesionreflects the degree to which individual membersof a group feel a part of, or "stuck to," theirsocial group(s). They believe that perceivedcohesion (a) relates more closely to other"subjective phenomena," such as loneliness,adjustment to a group, or suicide, than objectivemeasures of cohesion, and (b) mediates most ofobjectively measured cohesion's influences.

Bollen and Hoyle (1990, p. 482) definedperceived cohesion as "an individual's sense ofbelonging to a particular group and his or herfeelings of morale associated with membershipin the group." It presumably reflects an individu-al's appraisal of his or her relationship to aspecified group. Appraisal of one's groupbelongingess takes place both cognitively (basedon information the individual has accumulatedconcerning experiences within the group andwith group members) and affectively (based onfeelings about the aforementioned experiences).Morale presumably reflects a "global affectiveresponse" (combining positive and negativeemotional responses) from belonging to a group.On the basis of theories concerning the relation-ship between cognition and affect, the cognitiveand affective elements of perceived cohesion areexpected to be reciprocally related. At the

individual level, perceived cohesion reflects thegroup's role in its members' lives. Combininggroup members' perceptions provides a group-level perspective on cohesion and a "bottoms-up" viewpoint on individuals' roles in the group.

The perceived-cohesion construct does notmention attraction to the group or to groupmembers, with the important consequence that itapplies to small groups (at least those with somehistory of interaction) as well as to larger groupsand aggregates, including communities, organi-zations, and perhaps even societies. Thus,groups and social entities of quite different sizescan be compared on a common scale as regardstheir level of perceived cohesion. Nor is perceivedcohesion theoretically expected by its proponents torelate to sociometric or social network measures ofcohesion assessing the numbers and types ofgroup member interactions.

Bollen and Hoyle (1990) contended that (a)belongingness and morale are two primarydeterminants of perceived cohesion and (b) asense of belonging is "fundamental" to agroup's existence. In their words,

If individuals do not perceive themselves to bemembers of a group, it is difficult to understand howgroup norms, values, and other group characteristicsare likely to affect them Use of the term "group"implies some minimal sense of belonging on the part ofgroup members, otherwise the collection of individualsis an aggregate, (p. 484)

They also suggested that the concept ofbelongingness is relevant to self-categorizationtheory (discussed below).

Bollen and Hoyle (1990) developed a 6-item"perceived cohesion" scale (PCS). A sample"belongingness" item states "I feel a sense ofbelonging to ." A sample item assessingfeelings of morale is "I am enthusiastic about

." In the mid-1980s, they administeredthe PCS to two samples expected a priori todiffer in overall cohesion levels: (a) a sample ofjust over 100 students of various years at a smallnortheastern U.S. college known for its "strongschool spirit" and (b) residents of a mid-sizednortheastern U.S. city selected at random fromthe phone directory and diverse in terms of bothage and social class.

CFAs supported Bollen and Hoyle's (1990)two-correlated-factor model of perceived cohe-sion, as well as hypotheses of factorial invari-ance across the two groups and across middle-class and working-class segments of the

Page 12: Group Cohesion: From Field of Forces to Multidimensional Construct

18 DION

community sample, respectively. As predicted,perceived cohesion was indeed higher in thecollege sample than the city sample. Theinterfactor correlation was quite high in bothcases (.96 in the college sample and .92 in thecommunity sample). (Bollen and Hoyle did notthemselves test a one-factor model but providedthe covariance matrices, permitting me (oranyone else interested) to do so. The two-factormodel provided significantly better fit than theone-factor model for the city sample, whereasboth the one- and two-factor models weresupported for the college sample, with thedifference in chi-square between the two modelsjust failing to attain significance.)

Subsequently, Hoyle and Crawford (1994)explored some correlates of perceived cohesionin two undergraduate samples. In one case, PCSdimensions were correlated with social andpersonal outcomes. Belongingness tended tocorrelate more highly with social outcomes, andmorale tended to correlate more highly withpersonal outcomes. Both dimensions weremoderately correlated with having friends at theuniversity and being involved in universityactivities. The second study correlated PCSscales with various self-esteem dimensions andshowed that both belongingness and moralecorrelated somewhat more highly with socialself-esteem than with other types of self-esteem.

In sum, Bollen and Hoyle's (1990) constructof perceived cohesion is another recent exampleof a theory-driven approach to cohesion depart-ing from the prior conceptualization of cohesionas interpersonal attraction. Their model canapply to groups of widely varying sizes in whichmembers cannot or do not know all the membersas well as to groups in which members do knowone another or at least some, if not all, members.Its theoretical links to self-categorization andcognition-affect theories are additional "pluses"for their model, though the PCS remainsunderutilized as a cohesion measure thus far.

The belongingness component of the per-ceived-cohesion construct, as noted earlier,reflects a cognitively based sense of identifica-tion with certain groups stressed by socialidentity and self-categorization theorists. Themorale component, by contrast, reflects theaffective component of perceived cohesion,which would presumably relate positively toother measures of cohesion assessing interper-sonal attraction, attraction to the group, or both.

The belongingness and morale components ofperceived cohesion would seem to be differentaspects of social cohesion and reflect taskcohesion only indirectly, if at all. In any case,cohesion researchers can and should takeadvantage of the perceived-cohesion constructand the PCS's conceptual and empirical develop-ment by Bollen, Hoyle, and Crawford.

Self-Categorization Theory

A final example of a two-dimensional modelof cohesion is provided by self-categorizationtheory (SCT), as presented and elaborated byMichael Hogg, John Turner, and their col-leagues over the last 15 years. SCT stronglychallenges the traditional view of cohesion asinterpersonal attraction and makes a fundamen-tal distinction between interpersonal and groupprocesses, respectively. According to SCT>group behavior has characteristic features thatdifferentiate it from interpersonal behavior.People can categorize themselves at each ofthese different levels, though presumably not atthe same time. Categorization at the group levelleads to depersonalization (i.e., perceivingoneself and others not as unique individuals butrather in terms of the group's defining featuresor "prototype"), which underlies group phenom-ena, including cohesion. To SCT advocates,traditional research conceptualizing cohesion asinterpersonal attraction is reductionistic (i.e.,focused on the individual or interpersonal level)and basically out of touch with phenomenaoccurring at the group level.

The SCT analysis of cohesion distinguishestwo types of attraction within groups; aninterpersonal form of attraction called personalattraction (PA) and group-level attraction,termed social attraction (SA; Hogg, 1992). BothPA and SA are experienced as a positive ornegative attitude or feeling by one person foranother, and both may be present in a smallface-to-face group of interacting individuals.However, PA is allegedly idiosyncratic andgrounded in specific interpersonal relationships.SA, by contrast, is presumed to be a depersonal-ized form of liking that does not involvepersonal attraction but is based on members'prototypicality. That is to say, one's SAis basedon the extent to which members embodydefining essential features of the group asperceived by its members. SA is a generally

Page 13: Group Cohesion: From Field of Forces to Multidimensional Construct

SPECIAL ISSUE: GROUP COHESfON 19

positive attitude toward in-group members(because in-group prototypes are generallypositive) and a negative attitude toward out-group members (because out-group stereotypesare usually negative). According to SCT, SA isexpected to be intimately linked to other groupphenomena, whereas PA is not expected to be.

The SCT analysis of cohesion is nowsustained and supported by considerable evi-dence (see Hogg, 1992). Recent studies by Hoggand his colleagues show that when one's groupmembership is psychologically salient, likingfor others relates closely to their prototypicalityto the group, both in laboratory groups and inreal-world sport groups such as an Australianmen's football team (Hogg & Hardie, 1991) andwomen's netball teams (Hogg & Hains, 1996).

In the sports teams research, as an example(Hogg & Hains, 1996; Hogg & Hardie, 1991),PA was assessed by sociometric choices amongteam players as to which three members theywould select for a social activity outside theteam, such as going on a holiday, going to aconcert or a play, and the like. By contrast, SAwas indexed by sociometric choices of whichthree team players they would select if they weregoing to start up a new team. Because one wouldprobably select the best players for a new team,SA in these studies is most likely a variant oftask cohesion (a point acknowledged by Hogg &Hains, 1998). By contrast, PA would reflect atype of social or interpersonal cohesion unrelated tothe team's task performance and functioning.

Finally, prototypicality was measured byhaving the respondents rank team playerssociometrically in terms of the team's quintessen-tial characteristics, as they saw them, withoutthese being specifically defined. In both of thesesports team studies by Hogg and his associates(Hogg & Hains, 1996; Hogg & Hardie, 1991),prototypicality was associated more stronglywith SA than with PA. However, at least in thewomen's netball team study, and perhaps themen's football team study as well, PA and SAalso correlated positively with another.

SCT is perhaps the dominant social psycho-logical perspective on cohesion in the 1990s andhas reinvigorated cohesion research both concep-tually and empirically. Ironically, though, thetwo main critiques that SCT advocates haveleveled at research on cohesion as interpersonalattraction—(a) reliance on interpersonal attrac-tion measures to assess cohesion and (b) use of

the individual rather than the group as the unit ofanalysis—can, in turn, be directed at SCT itself,because sociometric measures taken from indi-viduals have been used to assess PA and SA(e.g., Hogg & Hains, 1996; Hogg & Hardie,1991) in SCT studies.

Likewise, despite SCT's emphasis on thegroup level, individuals are almost always usedas the unit of analysis in SCT research oncohesion. Even conducting statistical tests todetermine whether to use the group or theindividual as the unit of analysis has beenhonored more in the breach than in theobservance by SCT researchers. Because SCTfundamentally distinguishes between interper-sonal and group processes, advocates of thisviewpoint should obviously reflect the concep-tual distinction between group and individuallevels in their analyses and their measurement ofcohesion. Because the levels-of-analysis issue isof considerable importance to cohesion re-search, I discuss it further below.

Levels of Analysis

Cohesion research raises complex but impor-tant and interesting questions about the indi-vidual versus the group as the conceptual andstatistical levels of analysis. Conceptually, somesay cohesion is a property of the group as awhole rather than the individuals who composeit, although it is manifested at the individuallevel in forms such as attraction to the group, itsmembers, its tasks, or some combination ofthese. In general, the measurement and treat-ment of cohesion data should follow closelyfrom one's theoretical definition of cohesion,assuming that statistical analysis bears out one'stheoretical assumptions about the proper unit. Ifone's theory defines cohesion as a group-levelconcept, it should be measured at the grouplevel; similarly, if cohesion is construed as anindividual phenomenon such as interpersonalattraction, it should be assessed and treatedusing the individual as the unit of analysis, withthe proviso given above.

Thus far, psychologists have mainly assessedcohesion among individuals using cohesionscales or interpersonal-attraction measures. Evenif cohesion data are gathered at the individuallevel, they can be analyzed either at the group orindividual level. Group researchers now haveseveral well-established statistical procedures—

Page 14: Group Cohesion: From Field of Forces to Multidimensional Construct

20 DION

such as a nested analysis of variance in whichthe groups-within-treatment effect is testedagainst a liberal alpha of .20 or .30 (Anderson &Ager, 1978), calculating an intraclass correla-tion for a given measure reflecting the degree towhich variance among groups differs fromvariance within groups (Kenny & Judd, 1996;Kenny & La Voie, 1985), and a bootstrapprocedure in which the sampling distribution ofthe test statistic is empirically derived fromrepeated resamplings of the set of observations(Burlingame, Kircher, & Honts, 1994)—fordeciding whether the individual or the groupshould be taken as the unit of analysis for data atthe individual level. When the individual as theunit of analysis is justified statistically, analysisof individual-level data can clarify issues ofinterest to cohesion researchers, such as theindividual's experience of belongingness togroups and the impact of group membership onpeople's lives (see Hoyle & Crawford, 1994).

Sometimes, though, cohesion researchers usethe individual as the unit of analysis when theobservations are probably statistically depen-dent on one another, as a result of prior historyor interaction and influence among groupmembers. In such cases, the probability ofcommitting a Type I error is inflated and exceedsthe nominal alpha level to varying degrees. Suchinflation can be considerable and is greater forsmall groups (say, three persons) rather thanlarger ones (say, eight persons)—the formerbeing the type of group social psychologists aremore apt to study (Burlingame ct al., 1994;Kenny & Judd, 1986).

Using the individual as the unit of analysiswhen the group should be the proper unit alsocreates conceptual confusion. For cohesionresearch, the group level is as important toconsider as the individual level, because thesetwo levels inform one another, and eachprovides an important perspective. The group(or nomothetic) perspective may be especiallyimportant for designing interventions, because itmay be considerably more efficient, for ex-ample, to devise ways to reduce workloadrequirements for a group under a heavyworkload than to teach individual group mem-bers how to cope with a heavy workload (seeBliese & Halverson, 1996), when the group isthe proper unit of analysis. Finally, differentfindings may be obtained when groups versusindividuals are taken as the unit of analysis. In

their meta-analysis of the cohesion-perfor-mance relationship, Gully et al. (1995) foundthat correlations were notably stronger forcohesion studies that used the group rather thanthe individual as the unit of analysis and thatconfidence intervals for group versus individualeffect sizes did not even overlap.

As Burlingame et al. (1994, p. 499) remindedus: "Small group research is a challengingenterprise that requires a rich armamentarium ofanalytic tools." Fortunately, new sets of toolsallow us to consider both individual and thegroup as units of analysis simultaneously, asituation not unlike "having one's cake andeating it, too." One of these tools is within andbetween analyses (WABA), a procedure forforming and comparing between-unit and within-unit correlations (Bliese & Halverson, 1996). InWABA, the between-unit correlation is agroup-level correlation based on group meansweighted by group size. The within-unit correla-tion is based on the individual's relative positionwithin the group and holds group differencesconstant. With WABA, one infers group andindividual effects from a covarianee theorem.

In group-level analyses, aggregation of indi-vidual responses into a group score treats groupmembers as judges rating the environment.Therefore, it is important to demonstrate thatgroup members "agree" before claiming that aconstruct is a group-level variable (see Conway& Schaller, 1998; James, 1982; James, Dema-ree, & Wolf, 1984, 1992; Kozlowski & Hattrup,1992). Indeed, taking the group as the unit ofanalysis without evidence of within-groupagreement raises aggregation bias as a potentialproblem and plausible alternative explanation ofone's findings.

Unfortunately, few investigators bother toassess within-group agreement before taking thegroup as the unit of analysis, though there aresome notable exceptions and exemplars (e.g.,Bliese & Halverson, 1996; George & Bettenhau-sen, 1990). Tn Bliese and Halverson's (1996)analyses of data from companies of U.S. Armyunits, for example, within-group agreement wasassessed by two procedures: (a) calculating anintraclass correlation reflecting within-groupagreement and (b) a random-resampling proce-dure in which variances from actual groups werecompared with variances from numerous ran-domly formed "groups."

Page 15: Group Cohesion: From Field of Forces to Multidimensional Construct

SPECIAL ISSUE: GROUP COHESION 21

A second tool one can use to considermultiple levels of analysis is hierarchical linearmodeling (HLM), also known as multilevelmodeling (Kreft & de Leeuw, 1998). Within anorganization, individuals are nested within workgroups, work groups are nested within depart-ments, and departments are nested withinorganizations. To investigate these variouslevels, one must measure variables and con-structs describing each of these levels. Hierarchi-cal relationships arise when variables at onelevel of analysis influence, or are influenced by,variables at another level of analysis. Hierarchi-cal linear models explicitly recognize thatindividuals within a particular group may bemore similar to one another than individuals inother groups and, therefore, may not provideindependent observations. Also, these modelsallow one to investigate both lower and higherlevel unit variance in the outcome measurewhile maintaining the appropriate unit ofanalysis for the independent variables. There-fore, one can model both individual- andgroup-level variance in individual outcomeswhile utilizing individual predictors at theindividual level and group predictors at thegroup level. HLM, then, allows the investigationof relationships within a particular hierarchicallevel as well as relationships between or acrosshierarchical levels. HLM is apt to become acommon data analytic procedure in cohesionresearch, as well as social interaction research ingeneral (Kenny, 1996). Readers interested in theapplication of HLM to groups research maywish to consult several articles published in aspecial issue of Group Dynamics: Theory, Re-search, and Practice (Moritz & Watson, 1998;Nezlek&Zyzniewski, 1998; Pollack, 1998).

Conclusion

Cohesion research has at least some of theingredients necessary to thrive in the 21stcentury. The 1980s and 1990s witnessed theindependent emergence of several programs ofresearch on cohesion that are theory driven andmethodologically innovative in suggesting newmeasures of cohesion in diverse types of groups.As we enter its second century, cohesionresearch is now considerably more interestingand promising than it has been in a long time.

The conceptual distinction between taskcohesion and social cohesion that has emerged

independently from several models and lines ofresearch is an important milestone in cohesionresearch (Dion & Evans, 1992) and one whoseimportance seems to have a good deal ofsupport, if not consensus, from cohesion re-searchers today. The task-social distinction haseven been proposed as being a primary dimen-sion of cohesion that applies to most, if not all,groups (see Cota, Evans, Dion, Kilik, &Longman, 1995, and the discussion of theheuristic model below). Accordingly, anyoneconsidering cohesion as an independent ordependent variable (or both) in his or herresearch would do well to take it into account.Similarly, the vertical-horizontal cohesion dis-tinction reflecting members' perceptions ofleaders' competencies is apt to be another "pri-mary" dimension of cohesion relevant to manynatural groups, because most (if not all) smallgroups, including therapy groups, possess a leader-ship and influence structure among members.

On the measurement side, the situation isperhaps less sanguine and in need of systematicanalysis. We have no standard, off-the-shelfmeasure of cohesion in which we can havestrong confidence. Perhaps the best that can besaid is that we have some promising question-naire measures, such as the GEQ and the PCS,that require further exploration of their constructvalidity. The diversity of cohesion measures inthe literature also makes it difficult to comparestudies, though there are also advantages tohaving diverse ways to assess as rich andcomplex a construct as cohesion (see Dion &Evans, 1992). The PCS may qualify as a generaland broadly applicable measure of cohesion,one useful both for small, interacting groups andlarger ones in which members know some, butnot all, of the members.

At present we have little or no idea as to howdifferent putative measures of cohesion in ourcurrent toolbox relate to one another. Cohesionresearchers might include several cohesionmeasures where possible, taking care to ensurethat the sample size of groups is adequate toassess meaningfully, and with adequate statisti-cal power, the strength of any relationshipsamong the cohesion measures. For example,because the GEQ and the PCS both assessperceived cohesion by individual group mem-bers, it would be interesting to see how they andtheir subscales interrelate with one another.

Page 16: Group Cohesion: From Field of Forces to Multidimensional Construct

22 DION

Even more boldly, a multitxait-multimethodstudy of cohesion is long overdue, perhaps usingbehavioral (e.g., frequency of "we" statementsduring interaction) and questionnaire measuresof cohesion, as well as a different construct.

Even the cohesion-as-interpersonal-attractionperspective still remains to be empiricallyresolved, although the theoretical lines havebeen clearly drawn (e.g., by self-categorizationtheorists). In fact, interpersonal attraction innatural groups, such as members of a policeforce and football teams, has itself beensuggested as being multidimensional and consist-ing of three components: friendship, liking, andrespect (Segal, 1979). Segal (1979) found thatfriendship is more mutual than either liking orrespect. Liking and respect would appear toreflect social and task cohesion, respectively.Thus, a contemporary conceptualization ofcohesion as interpersonal attraction retains thesocial-task distinction and provides perhaps twotypes of social cohesion (viz., friendship andliking). Recent social perception research alsosuggests that liking persons may differ fromliking groups, in that the former depends onindividuals' personal desirability and the latteron similarity between self and group (Clement& Krueger, 1998). These contentions are clearlycongruent with self-categorization theory's dis-tinction between personal attraction (liking forpersons) and social attraction (based on proto-typicality of self to the group).

Yet another important challenge for cohesionresearch in the 21st century is to determinewhether one or more measures and conceptual-izations of cohesion can indeed apply to most, ifnot all, groups. One perspective is that "cohesive-ness . . . means different things for differentgroups with different goals and tasks."1 Fromthis perspective, it follows that no singlemeasure or conceptualization (except the neld-of-forces viewpoint?) could ever apply broadlybecause the construct changes for differentgroups (see footnote 1).

An alternative perspective was enunciatedrecently by Cota, Evans, et al. (1995), whoproposed a "new heuristic for cohesion."According to this heuristic perspective, there areprimary dimensions that would characterize thecohesion of most, if not all, groups, whereassecondary dimensions of cohesion would applyonly to specific types of groups. Four candidatesfor primary dimensions were proposed: (a & b)

the individual-group and social-task distinc-tions from Carron's hierarchical model, (c)normative views of group members, and (d)resistance of a group to disruptive forces. Risk-taking, vertical cohesion, and valued roles weresuggested as possible examples of secondarydimensions of cohesion, relevant for only specificgroups, such as therapy or sports groups.

Although I was among the coauthors of theaforementioned "new heuristic for cohesion,"the benefit of several years' hindsight and theopportunity for a personal perspective lead meto propose a somewhat different list of candi-dates for primary dimensions of cohesion.Bollen and Hoyle's (1990) contention thatbelongingness is fundamental to group cohesionis persuasive and qualifies it as a primarydimension. Also, as noted earlier, 1 believe thatvertical cohesion could well be a primarydimension, because most, if not all, groupsdifferentiate very early on between leader(s) andfollowers, even if at a rudimentary level. Forexample, in a therapy group the therapist playsthe leader role, and other members receivingtherapy are followers. Thus, my personal list ofprimary dimensions would be social cohesion,task cohesion, vertical cohesion, and belonging-ness. Resistance to disruptive forces fails tomake my personal list of primary dimensionsbecause it is perhaps more properly viewed as aconsequence rather than a cause of cohesion.2

Likewise, development of norms is perhapsmore a definition of a small group rather thancohesion per se.

The heuristic model of cohesion was anattempt to stimulate research into the structureof cohesion. Even if there are primary dimen-sions of cohesion that apply to, say, small groupsof interacting members, would they also applyto much larger groups, such as communities orcountries? Is cohesion identical in its causes andconsequences for large and small groups ornot?3 Clearly, these and other exciting andinteresting issues await us cohesion researchersin the 21st century.

11 am indebted to an anonymous reviewer for pointingthis out.

I am indebted to a second anonymous reviewer forpointing this out.

31 am indebted to Kurt Back for posing these searchingquestions.

Page 17: Group Cohesion: From Field of Forces to Multidimensional Construct

SPECIAL ISSUE: GROUP COHESION 23

References

Albert, R. S. (1953). Comments on the scientificfunction of the concept of cohesiveness. AmericanJournal of Sociology, 21, 231-234.

Allport, G. W., Vernon, P. E., & Lindzey, G. (1951).Study of values (Rev. ed.). Boston: HoughtonMifflin.

Anderson, L. R., & Ager, J. W. (1978). Analysis ofvariance in small group research. Personality andSocial Psychology Bulletin, 4, 341-346.

Back, K. (1951). Influence through social communica-tion. Journal of Abnormal and Social Psychology,46, 9-23.

Bales, R. F. (1950). Interaction process analysis: Amethod for the study of small groups. Cambridge,MA: Addison-Wesley.

Bales, R. F. (1999). Social interaction systems:Theory and measurement New Brunswick, NJ:Transaction.

Bales, R. F., & Cohen, S. P. (1979). SYMLOG: Asystem for the multiple level observation of groups.New York: Free Press.

Bernthal, P. R., & Insko, C. A. (1993). Cohesivenesswithout groupthink. Group & Organization Man-agement, 18, 66-87.

Bliese, P. D., & Halverson, R. R. (1996). Individualand nomothetic models of job stress: An examina-tion of work hours, cohesion, and well-being.Journal of Applied Social Psychology, 26, 1171-1189.

Bollen, K. A., & Hoyle, R. H. (1990). Perceivedcohesion: A conceptual and empirical examination.Social Forces, 69, 479-504.

Brawley, L. R., Carron, A. V., & Widmeyer, W. N.(1987). Assessing the cohesion of teams: Validityof the Group Environment Questionnaire. Journalof Sport Psychology, 9, 275-294.

Brawley, L. R., Carron, A. V., & Widmeyer, W. N.(1988). Exploring the relationship between cohe-sion and group resistance to disruption. Journal ofSport & Exercise Psychology, 10, 199-213.

Brawley, L. R., Carron, A. V, & Widmeyer, W. N.(1993). The influence of the group and itscohesiveness on perceptions of group goal-relatedvariables. Journal of Sport & Exercise Psychology,15, 245-260.

Burlingame, G. M, Kircher, J. C , & Honts, C. R.(1994). Analysis of variance versus bootstrapprocedures for analyzing dependent observationsin small group research. Small Group Research, 25,486-501.

Carron, A. V. (1988). Group dynamics in sport.London, Ontario, Canada: Spodym.

Carron, A. V., Brawley, L. R., & Widmeyer, W. N.(1985). The Group Environment Questionnaire.Unpublished manuscript, University of WesternOntario, London, Ontario, Canada.

Carron, A. V., Hausenblas, H. A., & Mack, D. (1996).Social influence and exercise: A meta-analysis.Journal of Sport & Exercise Psychology, 18, 1-16,

Carron, A. V., & Ramsay, M. C. (1994). Internalconsistency of the Group Environment Question-naire modified for university residence settings.Perceptual & Motor Skills, 79, 141-142.

Carron, A. V., & Spink, K. S. (1995). The groupsize-cohesion relationship in minimal groups.Small Group Research, 26, 86-105.

Carron, A. V., Widmeyer, W. N., & Brawley, L. R.(1985). The development of an instrument to assesscohesion in sport teams: The Group EnvironmentQuestionnaire. Journal of Sport Psychology, 7,244-266.

Carron, A. V., Widmeyer, W. N., & Brawley, L. R.(1988). Group cohesion and individual adherenceto physical activity. Journal of Sport & ExercisePsychology, 10, 127-138.

Cartwright, D. (1968). The nature of group cohesive-ness. In D. Cartwright & A. Zander (Eds.), Groupdynamics: Research and theory (3rd ed., pp.91-109). New York: Harper & Row.

Cattell, R. B. (1948). Concepts and methods in themeasurement of group syntality. PsychologicalReview, 55, 48-63.

Cattell, R. B., Saunders, D. R., & Stice, G. F. (1953).The dimensions of syntality in small groups.Human Relations, 6, 331-356.

Cattell, R. B., & Wispe, L. G. (1948). The dimensionsof syntality in small groups. Journal of SocialPsychology, 28, 57-78.

Clement, R. W., & Krueger, J. (1998). Liking personsversus liking groups: A dual-process hypothesis.European Journal of Social Psychology, 28,457-469.

Cohen, J. (1992). A power primer. PsychologicalBulletin, 112, 155-159.

Conway, L. G., IN, & Schaller, M. (1998). Methodsfor the measurement of consensual beliefs withingroups. Group Dynamics: Theory, Research, &Practice, 2, 241-252.

Cota, A. A., Dion, K. L., & Evans, C. R. (1993). Areexamination of the Gross Cohesion Scale.Educational and Psychological Measurement, 53,499-506.

Cota, A. A., Dion, K. L., & Evans, C. R. (1995). Anerror in Cota, Dion, and Evans (1993) produced bya GLS error in EQS Version 3.0. Educational andPsychological Measurement, 55, 684—686.

Cota, A. A., Evans, C. R., Dion, K. L., Kilik, L., &Longman, R. S. (1995). The structure of groupcohesion. Personality and Social PsychologyBulletin, 21, 572-580.

Cota, A. A., Longman, R. S., Evans, C. R., Dion,K. L., & Kilik, L. (1995). Using and misusingfactor analysis to explore group cohesion. Journalof Clinical Psychology, 51, 308-316.

Page 18: Group Cohesion: From Field of Forces to Multidimensional Construct

24 DION

Dion, K. L. (1979). Intergroup conflict and intragroupcohesiveness. In W. G. Austin & S. Worchel (Eds.).The social psychology of intergroup relations (pp.211-224). Pacific Grove, CA: Brooks/Cole.

Dion, K. L. (1990). Group morale. In R. Brown (Ed.),The Marshall Cavendish Encyclopedia of personalrelationships: Human behavior, how groups work(Vol. 15, pp. 1854-1861). Oxford, England:Cavendish.

Dion, K. L., & Evans, C. R. (1992). On cohesiveness:Reply to Keyton and other critics of the construct.Small Group Research, 23, 242-250.

Driedger, L. (1996). Multi-ethnic Canada: Identities& inequalities. Toronto, Ontario, Canada: OxfordUniversity Press.

Durkheim, E. (1951). Suicide (J. Spaulding & G.Simpson, Trans.). Glencoe, IL: Free Press. (Origi-nal work published 1897)

Dyce, J. A., & Cornell, J. (1996). Factorial validity ofthe Group Environment Questionnaire amongmusicians. Journal of Social Psychology, 136,263-264.

Eisman, B. (1959). Some operational definitions ofcohesiveness and their interactions. Human Rela-tions, 12, 183-189.

Evans, C. R., & Dion, K. L. (1991). Group cohesionand performance: A meta-analysis. Small GroupResearch, 22, 175-186.

Evans, N. J., & Jarvis, P. A. (1986). The GroupAttitude Scale: A measure of attraction to group.Small Group Behavior, 17, 203-216.

Festinger, L. (1950). Informal social communication.Psychological Review, 57, 271-282.

Festinger, L., Schachter, S., & Back, K. (1950). Socialpressures in informal groups: A study of humanfactors in housing. Stanford, CA: Stanford Univer-sity Press.

George, J. M. (1990). Personality, affect, andbehavior in groups. Journal of Applied Psychology,75, 107-116.

George, J. M., & Bettenhausen, K. (1990). Understand-ing prosocial behavior, sales performance, andturnover: A group-level analysis in a servicecontext. Journal of Applied Psychology, 75,698-709.

Golembiewski, R. T. (1962). The small group: Ananalysis of research concepts and operations.Chicago: University of Chicago Press.

Griffith, J. (1988). Measurement of group cohesion inU.S. Army units. Basic and Applied SocialPsychology, 9, 149-171.

Gross, E. F. (1957). An empirical study of theconcepts of cohesiveness and compatibility. Unpub-lished honor's thesis, Harvard University.

Gross, N., & Martin, W. E. (1952). On groupcohesiveness. American Journal of Sociology, 57,546-554.

Gully, S. M., Devine, D. J., & Whitney, D. J. (1995).A meta-analysis of cohesion and performance:Effects of level of analysis and task interdepen-dence. Small Group Research, 26, 497-520.

Hogg, M. A. (1992). The social psychology of groupcohesiveness: From attraction to social identity.London: Harvester Wheatsheaf.

Hogg, M. A., & Hains, S. C. (1996). Intergrouprelations and group solidarity: Effects of groupidentification and social beliefs on depersonalizedattraction. Journal of Personality and SocialPsychology, 70, 295-309.

Hogg, M. A., & Hains, S. C. (1998). Friendship andgroup identification: A new look at the role ofcohesiveness in groupthink. European Journal ofSocial Psychology, 28, 323-341.

Hogg, M. A., & Hardie, E. A. (1991). Socialattraction, personal attraction, and self-categoriza-tion: A field study. Personality and Social Psycho-logical Bulletin, 17, 175-180.

Hoyle, R. R , & Crawford, A. M. (1994). Use ofindividual-level data to investigate group phenom-ena. Small Group Research, 25, 464-485.

Hunt, M. (1985). Profiles of social research: Thescientific study of human interactions. New York:Russell Sage Foundation.

James, L. R. (1982). Aggregation bias in estimates ofperceptual agreement. Journal of Applied Psychol-ogy, 67, 219-229.

James, L. R., Demaree, R. J., & Wolf, G. (1984).Estimating within-group interrater reliability withand without response bias. Journal of AppliedPsychology, 69, 85-98.

James, L. R., Demaree, R. J., & Wolf, G. (1992). rwg:An assessment of within-group interrater agree-ment. Journal of Applied Psychology, 78, 306-309.

Janis, I. L. (1982). Groupthink: Psychological studiesof policy decisions and fiascos. Boston: Houghton-Mifflin.

Keller, R. T. (1983). Predicting absenteeism fromprior absenteeism, attitudinal factors, and nonattitu-dinal factors. Journal of Applied Psychology, 68,536-540.

Keller, R. T. (1986). Predictors of the performance ofproject groups in R&D organizations. Academy ofManagement Journal, 29, 715-726.

Kenny, D. A. (1996). The design and analysis ofsocial-interaction research. Annual Review ofPsychology, 47, 59-86.

Kenny, D. A.̂ & Judd, C. M. (1986). Consequences ofviolating the independence assumption in ANOVA.Psychological Bulletin, 99, 422-431.

Kenny, D. A., & Judd, C. M. (1996). A generalprocedure for the estimation of interdependence.Psychological Bulletin, 119, 138-148.

Kenny, D. A., & La Voie, L. (1985). Separatingindividual and group effects. Journal of Personal-ity and Social Psychology, 48, 339-348.

Page 19: Group Cohesion: From Field of Forces to Multidimensional Construct

SPECIAL ISSUE: GROUP COHESION 25

Ketchum, J. D. (1965). Ruhleben: A prison campsociety. Toronto, Ontario, Canada: University ofToronto Press.

Kline, R. B. (1998). Principles and practice ofstructural equation modeling. New York: GuilfordPress.

Kozlowski, S. W. J., & Hattrup, K. (1992). Adisagreement about within-group agreement: Dis-entangling issues of consistency versus consensus.Journal of Applied Psychology, 77, 161-167.

Kreft, I., & de Leeuw, J. (1998). Introducingmultilevel modeling. Thousand Oaks, CA: Sage.

Lewin, K., Lippitt, R., & White, R. K. (1939).Patterns of aggressive behavior in experimentallycreated "social climates." Journal of SocialPsychology, 10, 271-299.

Li, R, & Harmer, P. (1996). Confirmatory factoranalysis of the Group Environment Questionnairewith an intercollegiate sample. Journal of Sport &Exercise Psychology, 18,49-63.

Libo, L. M. (1953). Measuring group cohesiveness.Ann Arbor: Institute for Social Research, Univer-sity of Michigan.

Lippitt, R. (1940). An experimental study of authori-tarian and democratic group atmospheres. Univer-sity of Iowa Studies in Child Welfare, 16(3),43-195.

Lott, A. J., & Lott, B. E. (1965). Group cohesivenessas interpersonal attraction: A review of relation-ships with antecedent and consequent variables.Psychological Bulletin, 64; 259-309.

Lott, B. E. (1961). Group cohesiveness: A learningphenomenon. Journal of Social Psychology, 55,275-286.

Marrow, A. J. (1969). The practical theorist: The lifeand work of Kurt Lewin. New York: Basic Books.

Mikalachki, A. (1969). Group cohesion reconsidered:A study of blue collar work groups. Unpublisheddoctoral dissertation, School of Business Adminis-tration, University of Western Ontario, London,Ontario, Canada.

Moreno, J. L. (1934). Who shall survive? Washing-ton, DC: Nervous and Mental Diseases.

Moritz, S. E., & Watson, C. B. (1998). Levels ofanalysis issues in group psychology: Using efficacyas an example of a model. Group Dynamics:Theory, Research, and Practice, 2, 285-298.

Mudrack, P. E. (1989). Defining group cohesiveness:A legacy of confusion? Small Group Behavior, 20,37-49.

Mullen, B., Anthony, T., Salas, E., & Driskell, J. E.(1994). Group cohesiveness and quality of decisionmaking: An integration of tests of the groupthinkhypothesis. Small Group Research, 25, 189-204.

Mullen, B., & Copper, C. (1994). The relationbetween group cohesiveness and performance: Anintegration. Psychological Bulletin, 115, 210-227.

Nezlek, J. B., & Zyzniewski, L. (1998). Usinghierarchical linear modeling to analyze groupeddata. Group Dynamics: Theory, Research, andPractice, 2, 313-320.

Patnoe, S. (1988). A narrative history of experimentalsocial psychology: The Lewinian tradition. NewYork: Springer-Verlag.

Pollack, B. N. (1998). Hierarchical linear modelingand the "unit of analysis" problem: A solution foranalyzing responses of intact group members.Group Dynamics: Theory, Research, and Practice,2,299-312.

Prapavessis, H., & Carron, A. V. (1996). The effectsof group cohesion on competitive state anxiety.Journal of Sport & Exercise Psychology, 18,64-74.

Prapavessis, H., & Carron, A. V. (1997a). Cohesionand work output. Small Group Research, 28,294-301.

Prapavessis, H., & Carron, A. V. (1997b). Sacrifice,cohesion, and conformity to norms in sports teams.Group Dynamics: Theory, Research, and Practice,7,231-240.

Rindskopf, D., & Rose, T. (1988). Some theory andapplications of confirmatory second-order factoranalysis. Multivariate Behavioral Research, 23,51-67.

Schachter, S. (1951). Deviation, rejection, andcommunication. Journal of Abnormal and SocialPsychology, 46, 190-207.

Schutz, R., Eom, H. J., Smoll, F. L., & Smith, R. E.(1994). Examination of the factorial validity of theGroup Environment Questionnaire. Research Quar-terly for Exercise and Sport, 65, 226-236.

Segal, M. W. (1979). Varieties of interpersonalattraction and their interrelationships in naturalgroups. Social Psychology Quarterly, 42, 253-261.

Selvin, H. C , & Hagstrom, W. O. (1963). Theempirical classification of formal groups. Ameri-can Sociological Review, 28, 399-411.

Siebold. G. L. (1999). The evolution of measurementof cohesion. Military Psychology, 11, 5-26.

Spink, K. S., & Carron, A. V. (1994). Group cohesionand effects in exercise classes. Small GroupResearch, 25, 26-42.

Stein, A. A. (1976). Conflict and cohesion: A reviewof the literature. Journal of Conflict Resolution, 20,143-172.

Steiner, I. D. (1972). Group process and productivity.New York: Academic Press.

Stokes, J. P. (1983). Components of group cohesion:Intermember attraction, instrumental value, andrisk taking. Small Group Behavior, 14, 163-173.

Stryker, S. (1977). Developments in "two socialpsychologies": Toward an appreciation of mutualrelevance. Sociometry, 40, 145-160.

Page 20: Group Cohesion: From Field of Forces to Multidimensional Construct

26 DION

Tajfel, H., & Turner, J. (1979). An integrative theoryof intergroup conflict. In W. G. Austin & S.Worchel (Eds.), The social psychology of inter-group relations (pp. 33—47). Monterey, CA:Brooks/Cole.

Tziner, A. (1982a). Differential effects of groupcohesiveness types: A clarifying overview. SocialBehavior and Personality, 10, 227-239.

Tziner, A. (1982b). Group cohesiveness: A dynamicperspective. Social Behavior and Personality, JO,205-211.

Van Bergen, A., & Koekebakker, J. (1959). Groupcohesiveness in laboratory experiments. AdaPsychologica, 16, 81-98.

Wallace, E. R., IV (1977). The development ofFreud's ideas on social cohesion. Psychiatry, 40,232-241.

Wheeless, L. R., Wheeless, V. E., & Dickson-Markman, F. (1982). The relations among social

and task perceptions in small groups. Small GroupBehavior, 13, 373-384.

Yalom, I. D. (1995). The theory and practice ofpsychotherapy (4th ed.). New York: Basic Books.

Yalom, I. D., & Rand, K. (1966). Compatibility andcohesiveness in therapy groups. Archives ofGeneral Psychiatry, 15, 267-275.

Zaccaro, S. J. (1991). Nonequivalent associationsbetween forms of cohesiveness and group-relatedoutcomes: Evidence for multidimensionality. Jour-nal of Social Psychology, 131, 387-399.

Zaccaro, S. J., & Lowe, C. A. (1988). Cohesivenessand performance on an additive task: Evidence formultidimensionality. Journal of Social Psychology,128, 547-558.

Zaccaro, S. J., & McCoy, M. C. (1988). The effects oftask and interpersonal cohesiveness on perfor-mance of a disjunctive group task. Journal ofApplied Social Psychology, 18, 837-851.