A Validation Study of BELBINs TEAM ROLES

TEAM ROLES VALIDATION 121

© 2001 Psychology Press Ltd

EUROPEAN JOURNAL OF WORK AND ORGANIZATIONAL PSYCHOLOGY, 2001, 10 (2), 121–144

A validation study of Belbin’s team roles

S.G. FisherDepartment of Psychology, University of Strathclyde,

Glasgow, UK

T.A. HunterOxford Psychologists Press, Oxford, UK

W.D.K. MacrossonInterface Studies Unit, University of Strathclyde, Glasgow, UK

The Belbin team role preferences of the members of 55 teams were assessed bythree independent methods: (1) Cattell’s 16PF (Form 5) personality questionnaire,(2) video observation of a business simulation exercise and subsequent analysiswith a Belbin behavioural checklist, and (3) Saville and Holdsworth’sOccupational Personality Questionnaire. The 338 participants were drawn inapproximately equal measure from managerial and non-managerial levels fromequal numbers of manufacturing and public service organizations. A multitrait–multimethod correlation matrix derived from the data collected from theparticipants was employed to evaluate the convergent and discriminant validitiesof the Belbin team roles. Application of the Campbell and Fiske criteria to thematrix did not produce clear support for discriminant validity. Application of acorrelated uniqueness model in a confirmatory factor analysis showed the Belbinteam role model to be overparameterized and to lack both convergent anddiscriminant validity. Further modelling revealed that the Belbin team roles fiteasily into a “Big Five” five-factor personality framework.

Since its public appearance in 1981 in the book, Management Teams: Why TheySucceed or Fail, Belbin’s theory of team roles (Belbin, 1981) has been the focusof considerable attention both in industry (Senior, 1997) and academia (Broucek& Randell, 1996; Dulewicz, 1995; Fisher, Hunter, & Macrosson, 1998). In hisbook Belbin (1981, p. 132) proposed that the following five criteria must befulfilled for the construction of an effective team: (1) that each membercontributes to achieving objectives by performing a functional role (professional/

Requests for reprints should be addressed to Professor W.D.K. Macrosson, Interface StudiesUnit, Graham Hills Building, University of Strathclyde, 40 George Street, Glasgow G1 1QE, UK.Email: [email protected]

http://www.tandf.co.uk/journals/pp/1359432X.html DOI: 10.1080/13594320143000591

122 FISHER, HUNTER, MACROSSON

technical knowledge) and a team role, (2) that an optimal balance in otherfunctional and team roles is needed, depending on the team’s goals and tasks, (3)that team effectiveness depends on the extent to which members correctlyrecognize and adjust to the relative strengths, (4) that personality and mentalabilities fit members for some team roles and limit their ability to play others,and, (5) that a team can deploy its technical resources to best advantage onlywhen it has the range and balance of team roles to ensure efficient team work.The central concept linking these criteria was that of the team role, and the bookManagement Teams: Why they succeed or fail laid out the development anddefinitions of that concept. There, Belbin (1981) posited the necessity for eightteam roles to be deployed within a team for it to operate optimally. He coinedspecial names for the team roles, giving their characteristics as:

Completer-finisher: Painstaking, conscientious, anxious; searches out errors andomissions; delivers on time; perfectionism; obsessional behaviour.

Chairman: Mature, confident, a good chairperson; clarifies goals, promotesdecision making; delegates well; inclined to be lazy; takes credit for effort of ateam.

Company worker: Disciplined, reliable, conservative, and efficient; turns ideasinto practical action; adherence to the orthodox and proven; obstructingchange.

Monitor-evaluator: Sober, strategic, discerning; sees all options; judgesaccurately; scepticism with logic, cynicism without logic.

Plant: Creative, imaginative, unorthodox; solves difficult problems; preoccupiedwith ideas and neglects practical matters; strong ownership of ideas.

Resource investigator: Extrovert, enthusiastic, communicative; explores oppor-tunities; develops contacts; loses enthusiasm once initial excitement haspassed.

Shaper: Challenging, dynamic, thrives on pressure; has the drive and courage toovercome obstacles; a proneness to frustration and irritation; inability torecover situation with good humour or apology.

Team worker: Co-operative, mild, perceptive, and diplomatic; listens, builds,averts friction, calms the waters; indecision on crucial issues; avoidingsituations that may entail pressure.

(Later Belbin, 1993, altered the titles of two of his team roles; the Chairmanbecame the Co-ordinator, the Company worker the Implementer. His earliernomenclature has been adopted in this paper because of its more widespread usein industry.) The Belbin proposal was that the members of the team must becapable, amongst themselves, of discharging all of these eight team roles.Helpfully, in his formulation, Belbin (1981) acknowledged that an individualmay be able to operate effectively in more than one team role, thereby releasingan optimal team from any constraints on numbers. From his experiments and


work with teams Belbin (1993) also posited that mental ability, current valuesand motivations, field constraints, experience, and role learning as well aspersonality all have an effect on team behaviour, although he declined tocomment on the relative effect of each. However, the great importance thatBelbin, in practice, ascribed to the dimension of personality within team role isrevealed first, by his Self Perception Inventory published in his book (1981) andsecond, by his collaboration with personality test producers. The publishers ofthe 16PF (Cattell, Eber, & Tatsuoka, 1970) personality questionnaire and theOccupational Personality Questionnaire (Saville, Holdsworth, Nyfield, Cramp,& Mabey, 1992) both collaborated with Belbin (Dulewicz, 1995) at differenttimes to develop equations for calculating team role scores from their respectivetest data. Clearly, Belbin held the view that focusing on personality datasimplified the question of who to choose for team membership: “The extractionfrom psychometric test material of a disposition towards certain team rolesenables executives to develop clear expectations about the potential contributionof a candidate. In other words, complex data processed to appear in a simple formis often preferred to scores on multi-dimensional tests on which it is difficult totake an overview” (Belbin, 1993, p. 32).

The heart of the Belbin theory was simple: For optimal performance a teamshould contain individuals capable of operating in each of the eight team roles hehad defined. This was the key concept. So, if an organization was selectingcandidates for membership of a given team it should make every endeavour toensure the presence of all eight team roles. Additionally, and of considerablepractical import, access to these team roles was possible through the applicationof commercially available personality questionnaires and the relevant formulae.Indeed, the predictive power of the formulation was claimed to be veryconsiderable (Belbin, 1981). In a series of experiments Belbin, Aston, andMottram (1976) had shown how deviation from the ideal consistently yieldedteams that performed poorly. So, it was not surprising when industry wasdownsizing and the duties of managers were being taken over by teams, that theBelbin team role theory was enthusiastically adopted. Indeed, the extent to whichhis ideas on team roles have impacted on industry is betokened by the remarkableway in which the names originally given to these eight team roles have beenabsorbed into the everyday vocabularies of this generation of management. Somuch so that Roderick (1997a, b) has thought it worthwhile to lampoon them in atwo-part article on the dark side of team roles.

Although Belbin did not elaborate the theoretical foundations for his theory ofteam roles, its roots may be clearly seen in two major threads of classicalorganizational theory. First, in Bureaucratic Theory originally developed by MaxWeber (see Gerth & Mills’, 1958, translation) the need for specialization and theneed to match competence and job responsibility are made explicit. Belbin’steam role model echoes that need in that it identifies individuals’ particularbehavioural traits and matches them to specialized roles within the team. Weber


also recognized the need for orderly processing of work with special provision toensure that every task was accomplished. In the latter regard, Belbin’sCompleter-finisher role fits the need perfectly, whereas the need for orderlyprocessing is met in the Chairman and Company worker roles. However,Weber’s insistence on an unambiguous chain of command flies in the face of thevery nature of a team and is simply inappropriate in the true team environment.The second major thread of classical organizational theory upon which Belbin’stheory rests is to be found in the theories intimately associated with the humanrelations movement. Advocates of the human relations approach believe that it isbeneficial to create positive human relations within the organization (Mayo,1946; Miles, 1965). In that regard the function of Belbin’s Team worker role isclearly pivotal, the co-operative, perceptive and empathetic Team workerensuring not only that the interpersonal climate within the team is kept asharmonious as possible but that the Chairman is always kept aware of therelational undercurrents flowing within the team. Within the human relationsmovement, also, the importance of participation in the decision-makingprocesses in the organization has been recognized from the earliest times.Belbin’s team theory clearly draws on this, each of his team roles feeding into thesequence of events associated with the delivery of a project. For example, Guptaand Taube (1985) in a survey of project management theories reaching back intothe 1970s, when Belbin was conducting his team-building experiments, definedthe stages in project management as: Project selection; Planning; Definingresources needed; Scheduling; Implementing; Measuring progress; Reporting;and Maintaining human relations. Belbin’s team roles map comfortably ontoGupta and Taube’s stages. Shapers, Chairmen and Plants all have a major role toplay in the planning and defining phases; Resource investigators and Companyworkers will have found their métier in the Implementation stage; Monitor-evaluators and Completer-finishers are ideally suited to the Measuring progressand Reporting stages respectively; the Team worker is quietly working at Main-taining human relations. The bureaucratic theory of Weber and the humanrelations theory can also be seen as providing theoretical underpinning for theearlier team role model of Benne and Sheats (1948), a model not dissimilar to thatof Belbin. Indeed, the “group task roles” of Benne and Sheats (1948) can be seenas resting on these same theories. In this way, Belbin’s empirical formulationclearly links to a well-established base of theory.

In the light of the continuing adoption (Huczynski, 1997) of the Belbin teamrole typology the issue of the validity of the model assumes increasingimportance. Two validation studies (Broucek & Randell, 1996; Dulewicz, 1995)have been reported but both, in their own ways, have raised questions rather thanfurnished confirmation of the validity of Belbin’s team role model. In the first ofthe validation studies, Dulewicz (1995) evaluated team roles by using twopersonality questionnaires, the Cattell Sixteen Personality Factor Questionnaire(16PF: Cattell et al., 1970) and the Occupational Personality Questionnaire—Concept 5.2 (OPQ: Saville et al., 1992) and suggested that a comparison of the


factor structure of the team roles from each of the personality instruments and anexamination of the correlations between the team roles when measured by bothinstruments should provide evidence of construct validity of Belbin’s team rolemodel. His exploratory factor analysis yielded factor loadings that had little incommon and, additionally, failed to support Belbin’s proposed four pairings ofteam roles. Lack of support of construct validity may be inferred from this result.Helpfully, Dulewicz (1995) published the correlation matrices from which theexploratory factor analysis may be replicated. When the factor analysis isreplicated and the relevant anti-image correlation matrices and indices derivedfrom it calculated, the appropriateness of conducting a factor analysis is seen asbeing highly questionable. The Kaiser (1981) measure of sampling adequacy forfactor analysis is found to be .43 and .44 for the 16PF and OPQ data respectively,both of which, in Kaiser’s terms, are “unacceptable”. The usefulness of thepattern of factor loadings in establishing construct validity in this instance may,at best, be considered doubtful. The second set of data that Dulewicz suggestscould help establish construct validity is the matrix of correlations between the16PF and OPQ derived team roles. First, the validity diagonal in that matrixshows the average correlation to be .35; second, of the 56 off-diagonalcorrelations 22 are significant, but the mean value of these 22 correlations is amere .31. These data are advanced as evidence for “inter-method/equivalent formreliability , and construct validity of seven of the eight team roles” (Dulewicz,1995, p. 91). The authors of this paper take a different view and suggest that anaverage correlation of .35 signals poor “inter-method/equivalent form reliability”and that 39% of off-diagonal correlations being statistically significant andhaving a mean value comparable to the validity diagonal itself indicates a lack ofdiscriminant validity.

In the second, more recent, study of construct validity, Broucek and Randell(1996), used the five-factor model as operationalized by the NEO-PI/NEO-PI-Rquestionnaire (Costa & McCrae, 1992) as a standard against which to evaluatethe Belbin team roles as measured by both the Self Perception Inventory andBelbin’s Observer Assessment instrument. They concluded on the basis of thecorrelations between the NEO-PI/NEO-PI-R and Self Perception Inventory thatthere was no evidence to support claims for either convergent or discriminantvalidity. Solely on the basis of the well-established inadequacy of the psycho-metric properties of the Self Perception Inventory the authors take no issue withthat conclusion. However, the use of the NEO-PI/NEO-PI-R questionnaire toexamine the discriminant validity of Belbin’s Observer Assessment instrumentwhich, unlike the Self Perception Inventory had not been the subject of academicenquiry prior to the work of Broucek and Randell (1996), depends on theorthogonality of the dimensions of personality in the five-factor model asoperationalized in that questionnaire. Broucek and Randell (1996, p. 391) write,“These dimensions are orthogonal to one another …”, but inspection of the NEO-PI/NEO-PI-R Professional Manual (Costa & McCrae, 1992, p. 100) shows that anumber of substantial correlations do exist between them. Clearly, the five


factors are not orthogonal to one another. Any conclusion regarding the dis-criminant validity of the Belbin’s Observer Assessment instrument must,therefore, have a corresponding element of uncertainty attached to it.

In view of these uncertainties and an earlier examination of the structure ofBelbin’s team roles (Fisher et al., 1998) the time appeared ripe to adopt analternative approach to the issues of the validity and structure of the team rolemodel. The method chosen for the enquiry employed the well-established(Marsh, 1989) multitrait–multimethod matrix analysis followed by confirmatoryfactor analysis. The multitrait–multimethod matrix was first proposed byCampbell and Fiske (1959) as a way of establishing the convergent anddiscriminant validity of a given construct. They proposed that the traits underscrutiny, of which there should be three or more preferably, be measured by atleast three different methods. In this study three measures were employed: Twowere derived from separate personality questionnaires, the 16PF and the OPQ,and the third from a video assessment technique devised specifically for thisenquiry. Thus, the team roles yielded eight traits and the 16PF, OPQ, and videoassessment three methods for an 11 × 11 multitrait–multimethod matrix. The SelfPerception Inventory was not included in the multitrait–multimethod matrixportfolio of measures on account of the overwhelming evidence (Fisher,Macrosson, & Sharp, 1996; Furnham, Steele, & Pendleton, 1993) accumulated inregard to its unsatisfactory psychometric properties. Adoption of the multitrait–multimethod matrix methodology avoided issues concerning the best personalityquestionnaire to employ or any assumptions regarding the underlying structure ofthe model. Thus, the approach taken in this study was deliberately conservative.By avoiding the much criticized Team Roles Self Perception Inventory andadhering to the personality questionnaire algorithms derived from Belbin wesought to keep on firm ground, noting that Dulewicz (1995, p. 94) had, in regardto these algorithms, indicated that “it seems reasonable to contend that these16PF equations form a benchmark for further research” (authors’ italics). Wealso employed our video observational technique, in preference to Belbin’sObserver Assessment report form which is only completed by fellow workers, inan effort to achieve reliability and impartiality. To achieve comparability withBelbin’s subjects we sought only volunteers who were in employment and hadprevious experience of working in groups or teams. We also recruited a suf-ficiently large sample to obtain statistical significance for small values of correla-tion coefficients thereby offering hope of detecting effects other than gross ones.

METHOD

Subjects

A total of seven prominent multinational manufacturing organizations and twolarge indigenous UK firms, all with plants located in Scotland, provided 14 teams(44 males, mean age = 36 years, SD = 8.4; 20 females, mean age = 30 years, SD


= 6.5) comprising managerial personnel and 14 teams (107 males, mean age = 39years, SD = 10.5; 11 females, mean age = 27 years, SD = 6.3) comprising non-managerial personnel. Public service organizations such as local district councilsand NHS trusts provided 14 teams (33 males, mean age = 33 years, SD = 9.4; 81females, mean age = 41 years, SD = 10.5) comprising managerial and 13 teams (3males, mean age = 29 yrs, SD = 6.7; 39 females, mean age = 36 years, SD = 10.9)non-managerial personnel. Thus, a total of 55 teams comprising 187 male and151 female volunteer subjects, mean age 38.1 (SD = 9.9) and 37.2 (SD = 10.9)years respectively, were recruited.

Questionnaires

16PF 5. The 16 Personality Factor Questionnaire Form 5 (16PF5) publishedby the Institute of Personality and Ability Testing and distributed in the UK byNFER-NELSON Ltd is a personality questionnaire that measures 16 primaryaspects of adult personality. The questionnaire was administered to each indi-vidual participating in the study and a raw score for each personality factor foreach participant was derived. The raw scores for each participant were insertedinto the formulae shown in Table 1. Sten scores for each team role were thenderived by inserting the raw team role scores into the norm table (not shown inthis paper) provided by the publishers, NFER-NELSON Ltd.

Towards the end of the 1970s, that is in the later stages of his studies, Belbin(personal communication) pinpointed individuals who had played crucial roles,as recorded by observers, in the business simulation exercises and found arelationship between psychometric test scores and observed team roles. Therelationship was uncovered by comparing the 16PF test scores of teams which

TABLE 1Formulae for the derivation of team role scores

from 16PF5 raw data

Team role Equation

Completer-finisher –C+G+O+Q3+Q4Chairman –A+C+2E+F+2G–2L–M+2Q3Company worker 2G–2I–2L–2M–Q1+Q3Monitor-evaluator –F+NPlant –2A+E-2F+H+2I+M–N+Q1+2Q2Resource investigator 2C+F+H-L+M+Q1Shaper –C+3E+F–G+H–I+2O+2Q4Team worker A–E–L–Q2

16PF5 Personality in Practice by Wendy Lord © NFER-NELSON2000. Adapted by permission of the Publishers, The NFER-NELSONPublishing Company Limited, Darville House, 2 Oxford Road East,Windsor, Berkshire SL4 1DF, England. All rights reserved.


were successful in the business simulation exercise with the test scores of theunsuccessful teams. On the bases of these observations Belbin produced hisequations as a set of handwritten tables (Berry, 1995); the equations are now inthe open literature (Fisher, Hunter, & Macrosson, 2000). There is no evidenceextant to indicate that these equations were derived by linear regression or anysimilar procedure. Recourse to mathematical techniques appears not to have beennecessary since Berry (1995) has been able to demonstrate plainly that the teamrole equations can be deduced from the descriptions given in Belbin’s first book(1981). However, one team role equation was an exception. Belbin discoveredthat the cluster of distinguishing characteristics of the Plant was none other thanCattell’s formulae for Creative Disposition; he “adopted it as it stood” (1981,p. 33). At the start of his work on team roles Belbin (personal communication)had been using the 1962 edition (Form A) of the 16PF but when that edition wassuperseded in 1969 he found the new form to be a less reliable predictor. At thatjuncture Belbin (personal communication) started to give more attention toobserver assessments than self reporting. In spite of the experience with the 1969edition of the 16PF, Belbin and the 16PF publishers, NFER-NELSON Ltd (Life,1993, personal communication) produced a set of equations (Dulewicz, 1995) forthe calculation of team roles. The 1969 version of the 16PF has now beensuperseded by the 1993 version, the 16PF5. The final stage in the evolution of theteam role equations was the substitution of the 1969 version primary factors bythe appropriate admixture of 16PF5 primary factors, moderated and informed byBelbin and NFER-NELSON Ltd (Life, 1993, personal communication). Theoutcome of this latter collaboration is the equations shown in Table 1; these arethe equations used in this investigation.

OPQ. The Occupational Personality Questionnaire (OPQ) Concept 5.2published by Saville and Holdsworth UK Ltd is designed to measure 31 separateaspects of an individual’s behaviour, interests, and personality characteristics in awork context (Saville et al., 1992). The questionnaire was administered to eachindividual participating in the study and a raw score for each personality factorfor each participant was derived. Standardized (Sten) scores for the 30 factorswere derived through the use of the appropriate commercially available normtables. The Sten scores were then converted into the eight team roles usingproprietary software that employs an algorithm derived in co-operation withBelbin and norms derived from over 1000 UK managers (Dulewicz, 1995). Theequations on which the algorithm is based were generated in three steps: first,conceptual descriptions were generated from the research literature; second,from these conceptual descriptions predictive equations were defined; third,early versions of the equations were submitted to Dr Belbin who added his ownviews and amendments, which were generally taken into account in the finalequations (Mabey & Hunter, 1986). However, no information on the algorithms,


norms, or team role equations, comparable to that disclosed by NFER-NELSONLtd, has been published in the open literature by Saville and Holdsworth UK Ltd.

Business simulation exercise. “Toxic Waste” is a commercially availableteam-working exercise obtained from Chadwick Rees Consultancy Ltd, England.It is an exercise that had been developed specifically to allow various aspects ofteam-working styles and behaviours to emerge, and was selected for this researchbecause no specialized business or technical knowledge was required for itscompletion. It also imposes few rules or any hierarchy on the team prior to thestart of the exercise. In advance of conducting the main investigation, a trial runof “Toxic Waste” was made with a team comprising volunteers drawn fromuniversity staff and students. From the lessons learned from the trial a minormodification was made to the exercise which helped bring out the characteristicsof the least assertive or involved team member. The modification to the “ToxicWaste” exercise was nothing other than addition of the requirement that the teamproduce a tender for the contract to collect the waste.

On a day agreed between the employing organization and the researcher(Hunter), one of the 55 teams who volunteered for the investigation met in aconference room on company premises or in a local hotel. After completing the16PF and OPQ personality questionnaires, administered by the researcher, theteam was invited to participate in the modified form of “Toxic Waste”. Theexercise, which has no time limit, usually took each team about 3 hours tocomplete. The entire business simulation exercise was recorded on videotape.This procedure was repeated for each of the 55 teams.

Video observation. All of Belbin’s writings were combed for adjectives anddescriptive statements relating to the eight team roles. Each group of adjectivesand descriptive statements were clustered under each team roles and then refinedand reduced to about 7–10 items that captured the essence of each team roles’scharacteristic behaviour. These groups of items, shown in Table 2, were used todefine the team roles and constituted our Belbin Behavioural Checklist, whichprovided the researcher with “markers” for specific team role behaviours. Sincethe business simulation exercises for all 55 teams had been recorded onvideotape, by replaying the tape time and again, the behaviour of every individualoperating in a team context could be assessed at leisure. The video recordings ofthe business simulation exercise were played repeatedly and the researcherfocused on each team member in turn and awarded the team member one point onevery occasion that the team member displayed any of the behaviours describedin the Belbin Behavioural Checklist. On completion of the event recording,summation of the points yielded a team role score for each subject.

In the pilot study in which a team comprising volunteers drawn fromuniversity staff and students completed “Toxic Waste”, the researcher’s scoring


TABLE 2Belbin behavioural checklist

CF Completer- finisher Thorough about a task; attentive to all details; finishes things;reluctant to let go until complete; nags others to finish on time;perfectionist; plans so that nothing gets overlooked.

CH Chairman Delegates tasks and rallies team; makes firm decisions usuallyhaving consulted others; adaptive to changes; committed to goalsand sees to them being met; uses all team members making allplay a role; draws out the potential from all members; recognizestalent and uses it.

CW Company worker Orderly and precise; sticks to rules; doesn’t like change;organizes plans and action; disciplined in approach; systematic inapproach; tackles any tasks.

ME Monitor- evaluator Asks for all information; unenthusiastic and impartial aboutideas; slow to make a decision; likes to think based on all facts;negative about plans.

RI Resource investigator Negotiates and liaises with outsiders; asks questions from others;opportunistic; thinks on feet; picks up on the ideas of others; issociable; inquisitive and curious; enthusiastic about tasks at thebeginning.

PL Plant Thinks things through; tackles problems usually on own;proposes ideas and solutions; creative and imaginative; dominantwith ideas.

SH Shaper Active in the need to achieve something; restless; likes to workwithout planning; challenges others or indecision in a team;argues; disturbs the balance; goads action from others;opportunist, often negative; impatient and often frustrated.

TW Team worker Listens to others and supports them; works with the awkwardpeople; not forceful or demanding; diplomatic and balancing;averts conflict; communicates well with others.

of team behaviour on the video recording was correlated with that of an educated,managerial level, lay person. Inter-observer reliabilities of between .7 and .9were obtained over the range of team roles, the Monitor-evaluator role giving thepoorest, the Resource investigator the highest reliability respectively.

Analysis

For each participant the experimental protocol yielded 24 data points: three teamrole scores, one each from the 16PF questionnaire, the OPQ questionnaire andthe video observations, for each of the eight Belbin team roles. Complete sets of


data were obtained for 331 out of the 338 participants. Two commerciallyavailable software packages were used to process the data, SPSS for Windows(release 6.1.3) for the calculation of correlation coefficients, and AMOS (release3.6) for the confirmatory factor analyses. The initial step in the analysis of thedata was the calculation of the multitrait–multimethod correlation matrix. Thetraits are the eight Belbin team roles, the methods the 16PF questionnaire, theOPQ questionnaire, and the video observation.

RESULTS AND ANALYSIS OF THE MULTITRAIT–MULTIMETHOD MATRIX

As indicated earlier, Campbell and Fiske’s (1959) approach to the establishing ofconvergent and discriminant validity of a given construct rests on the analysis ofa multitrait–multimethod matrix of Pearson’s correlations. The multitrait–multimethod matrix obtained in this study is shown in Table 3. All correlations inthe validity diagonal, shown in bold in the heterotrait–heteromethod blocks(encased in lines in Table 3), were significant at p £ .05; the heterotrait–monomethod correlation triangles are in italics.

Campbell–Fiske criteria

In their landmark paper, Campbell and Fiske (1959) suggested four criteria bywhich convergent and discriminant validity might be evaluated from a multitrait–multimethod matrix; these criteria were re-expressed by Marsh and Hocevar(1983) and Marsh (1989) as follows:

� Convergent validity criterion: The convergent validity coefficients shouldbe statistically significant and sufficiently different from zero to warrantfurther examination of validity.

� Discriminant validity criteria:(i) The convergent validities should be higher than the correlations

between different traits assessed by different methods (i.e., theconvergent validity should be higher that the values lying in its columnand row in the heterotrait–heteromethod triangles).

(ii) The convergent validities should be higher than correlations betweendifferent traits assessed by the same method.

(iii) The pattern of the correlations should be similar for each of thedifferent methods.

In Table 3 the validity diagonals, shown in bold face, were found to bestatistically significant and different from zero (mean = .48, SD = .10; 99%confidence interval, t(.01, 24) = 2.81, upper and lower limits, .54 and .42respectively). If the data have failed to meet this first criterion, which indicatesconvergent validity being approached, it would have been hard to avoid the


132

TA

BLE

3M

ult

itra

it–m

ult

imet

ho

d P

ears

on

co

rrel

atio

n m

atri

x fo

r B

elb

in te

am r

ole

s (N

= 3

31)

16P

FV

ideo

OP

Q

CF

CH

CW

ME

PL

RI

SHT

WC

FC

HC

WM

EP

LR

ISH

TW

CF

CH

CW

ME

PL

RI

SHT

W

CF

CH

06C

W16

711

ME

33–1

913

6P

L11

–25

–52

20P

RI

–60

32–1

5–6

1–0

2F

SH

36–0

5–2

3–2

506

09T

W–2

610

06–4

2–4

637

–29

CF

47–0

302

3511

–43

03–1

5C

H–1

247

18–2

4–0

340

1007

–14

CW

0827

4426

–17

–19

–24

–11

2207

VM

E23

–15

–01

3713

–37

–02

–24

38–1

622

iP

L–0

5–3

0–4

602

4708

13–2

001

–07

–13

17d

RI

–38

16–0

8–4

9–0

659

1519

–30

45–1

5–3

101

eS

H06

–08

–23

–16

1315

58–2

804

23–1

8–0

216

31o

TW

–09

–16

–14

–12

–12

01–2

451

07–2

0–1

3–0

9–0

610

–29

CF

4838

3928

–09

–30

–04

–15

4404

3121

–21

–28

–06

–16

CH

–27

4610

–35

–04

6314

17–2

259

08–1

7–0

137

20–2

210

CW

–05

5948

08–1

315

–19

–08

0640

4408

–22

06–0

6–2

861

60O

ME

2814

1725

09–2

705

–31

18–0

222

4008

–24

–06

–26

4300

35P

PL

–16

–06

–37

–21

3539

30–0

7–1

521

–15

0345

2330

–22

–16

4301

22Q

RI

–45

3802

–64

–16

7920

32–4

149

–11

–35

–02

5924

–10

–17

8129

–22

35S

H05

15–1

3–2

413

3057

–27

–16

29–1

2–0

919

2155

–46

0243

0707

4741

TW

08–0

4–0

8–1

9–1

202

–21

5412

–08

–09

–10

–10

–05

–30

5004

03–1

4–1

3–1

404

–54

CF

=F

inis

her;

CH

=C

o-or

dina

tor;

CW

=Im

plem

ente

r; M

E=

Mon

itor

eva

luat

or; P

L=

Pla

nt; R

I=R

esou

rce

inve

stig

ator

; SH

=S

hape

r; T

W=

Tea

m w

orke

r; d

ecim

al p

oint

som

itte

d; v

alid

ity

diag

onal

s in

bol

d; h

eter

otra

it–m

onom

etho

d da

ta tr

iang

les

in it

alic

s.


conclusion that the different methods had been by measuring differentconstructs; a lack of convergent validity in at least one method would have beenindicated.

In regard to the first of the discriminant validity criteria, inspection of Table 3reveals that 161 of the 168 off-diagonal values in the three heterotrait–heteromethod blocks are smaller than the convergent validities, implyingagreement on any one trait is relatively independent from agreement on othertraits, thereby giving support for this first criterion of discriminant validity.

The second criterion regarding discriminant validity requires that theconvergent validities be higher than correlations between different team rolesassessed by the same method. This requires that each value in the validitydiagonals be compared with all other correlations in the same row and samecolumn in the corresponding triangular heterotrait–monomethod submatrices(Marsh, 1990). For the three sets of comparisons involving 168 comparisons, 17correlations in the heterotrait–monomethod triangles were seen to be higher thantheir corresponding convergent validities. The mean convergent validity (.49) ishigher than the mean correlation among scales for the 16PF (.04), video (.08),and the OPQ (.15) methods. Most of the 17 failures originated in the OPQmonomethod submatrix These data were interpreted as lending support to overalldiscriminant validity since a method effect would have resulted in similar valuesfor correlations in the validity diagonals and correlations in the heterotrait–monomethod triangles.

The third criterion for discriminant validity set by Campbell and Fiske (1959)required that the pattern of correlations between different traits (team roles in thisstudy) be similar for the same and different methods. In other words, the patternof trait interrelationship should be visible in all of the heterotrait triangles of boththe monomethod and heteromethod blocks. As indicated by Marsh (1989, 1990),assuming that there are significant correlations among traits, satisfaction of thiscriterion would suggest that true trait correlations are independent of the methodof assessment, whereas failure would suggest that the observed correlations aredifferently affected by method effects. Inspection of the matrix did not revealclearly the hypothesized pattern and so this third criterion could not be declaredfirmly to have been met.

Difficulty in meeting this final criterion for discriminant validity has beenencountered from the very start of multitrait–multimethod matrix analyses,Campbell and Fiske (1959) having noted that a failure to meet this criterion wasencountered in many of the studies that they had examined. Part of the difficultyexperienced in meeting this criterion must reside in the fact that any conclusionbased on a visual inspection of a sizeable matrix in the absence of any quantifiedmetric cannot be other than arbitrary and subject to the desires, conscious orotherwise, and the blindness of the investigator. Thus, for the size of the matrix inthis study it was deemed prudent to disregard this third criterion for discriminantvalidity. In summary, these preliminary results do appear to provide support for


the convergent and discriminant validity of the team role constructs, although thesupport for discriminant is not as strong as that for convergent validity.

As indicated by Marsh (1989), Campbell and Fiske (1959, p. 85) implied anadditional criterion in stating that “the presence of method variance is indicatedby the difference in level of correlation between parallel values of themonomethod block and the heteromethod block, assuming comparablereliabilities among the tests”. Following the procedure given by Marsh (1989)each of the 84 heterotrait–monomethod correlations was compared with theappropriate four heterotrait–heteromethod correlations. For 210 of these 336comparisons heterotrait–monomethod correlations were larger than theheterotrait–heteromethod correlations, thereby suggesting the presence ofmodest method effects.

Because of the importance of establishing the convergent and discriminantvalidity of any given construct and the elegance of the Campbell and Fiske(1959) approach to this problem, many investigators have, down the decades,suggested novel approaches to the analysis of the multitrait–multimethod matrix,for example the three-factor unreplicated analysis of variance as described byKavanagh, MacKinney, and Wolins (1971). However, the development ofconfirmatory factor analysis and the widespread availability of user-friendlysoftware has transformed the analysis of the multitrait–multimethod matrix.Confirmatory factor analysis is now the preferred approach (Marsh, 1990).

Confirmatory factor analysis

Multitrait–multimethod matrices, like other correlation matrices, can be factoranalysed to infer underlying constructs. Factors defined by different measures ofthe same trait are used to infer trait effects, whereas factors defined by differentmeasures assessed with the same method are used to infer method effects (Marsh,1990). With confirmatory factor analysis the investigator can define models thatposit a priori either trait or method factors, or both, and test the ability of suchmodels to fit the data. In the general multitrait–multimethod model adapted fromJöreskog (1974) there are at least three traits (T = 3) and three methods (M = 3),the (T × M) measured variables are used to infer the (T + M) a priori factors; eachmeasured variable loads on one trait and one method factor but is constrained soas not to load on any other factors; and, finally, the correlations among traitfactors and among method factors are freely estimated, but correlations betweentrait and method factors are fixed to be zero (Marsh, 1989). These constraintsallow the decomposition of variance into additive trait, method and errorcomponents. In Widaman’s (1985) taxonomy of structural models for multitrait–multimethod data this structure is labelled Model 3C.

The confirmatory factor analysis models for multitrait–multimethod matricescan be defined in terms of three design matrices (1) a (T × M) by (T + M) matrixof factor loadings, (2) a (T + M) by (T + M) matrix of factor variances and


covariances , and (3) a (T × M) by (T × M) matrix of unique variances andcovariances; in the terminology of LISREL (Jöreskog & Sörbom, 1981) thesematrices are labelled L y, Y , and Q respectively. This model is particularlyattractive in that its structure directly corresponds to Campbell and Fiske’s(1959) original conceptualization of the multitrait–multimethod matrix.Unfortunately , estimation problems with the model are legion, out-of-rangeestimates, negative unique variances, and non-convergence being amongstfrequently encountered difficulties. In the face of these “virtually universalestimation problems” Kenny and Kashy (1992, p. 166) argued for the adoption ofa correlated uniqueness model, first proposed by Kenny (1979) and subsequentlychampioned by Marsh (1990). (The path diagram of this model is shown inFigure 1 for the case of five traits and three methods; the rectangles representobserved variables, the large ellipses the latent variables, and the small ellipsesthe specific error or unique variables. The arrows indicate the dependenciesbetween the variables.) The correlated uniqueness model specifies only traitfactors; no method factors are created. Instead, the disturbances, or uniquefactors, are allowed to be correlated across measures of the same method, asshown in Figure 1. Despite this constraint it is possible to assess convergent anddiscriminant validities as in traditional confirmatory factor analysis, which usesthe Widaman (1985) Model 3C.

Fitting the model

An attempt to confirm the convergent and discriminant validity of the Belbinmodel by fitting a correlated uniqueness model containing eight traits and threemethods to the multitrait–multimethod data shown in Table 3 was unsuccessful;the estimated correlation matrix was nonpositive definite. In a paper onnonpositive definite matrices in structural modelling, Wothke (1993) pointed tothe dangers of overparameterization, indicating that the number of parametersthat can reasonably be included in a structural model is partly determined by thedegree of systematic information contained in the data. As noted earlier, ameasure of sampling adequacy (Kaiser, 1981) was calculated for each of thecorrelation matrices for the Belbin team roles derived from the 16PF and OPQpersonality tests by Dulewicz (1995); these measures indicated that correlationsbetween pairs of variables cannot be explained by other variables, thus lendingweight to the suspicion that overparameterization may lie at the root of thenonpositive definite correlation matrix difficulty. On the basis of this pointer,pairs of team roles were combined by averaging team role scores in an effort toreduce the number of traits in the correlated uniqueness model. This appeared areasonable stratagem, Belbin (1981) himself having noted the intrinsic pairrelatedness of some of the eight individual team roles; he had suggested a 4 × 2taxonomy of Negotiators (Resource investigators and Team workers), Manager-workers (Company workers and Completer-finishers), Intellectuals (Monitor-


Figure 1. Correlated uniqueness model for five traits and three methods.

17

evaluators and Plants), and Team leaders (Chairmen and Shapers). So, as a firststep in a progressive reduction of the number of team roles, the scores of one pairof team roles were averaged giving six individual and one paired team role, a totalof seven traits; in the next step in the process of reduction a further pair of teamroles were combined to give four individual and two paired team roles, a total ofsix traits; this process was continued one further cycle. The correlated uniquenessmodel was, in turn, fitted to the reduced data sets thereby derived. Freedom fromnonpositive definite correlation matrices was not encountered until six team roleshad been put into three pairs, thereby yielding five traits, namely, three paired andtwo unpaired team roles. Some of the groupings used in the model fitting workare shown in Table 4.


The fit of the five trait, three method correlated uniqueness model, Figure 1, toeach set of data was evaluated on three measures. The first of these measures wasthe ratio of the minimum sample discrepancy function (Browne, 1982, 1984) tothe degrees of freedom; it has been held to be a good indicator (Marsh &Hocevar, 1985) with values of between 2 and 5 being taken to signal a reasonablefit. Wheaton, Muthén, Alwin, and Summers (1977) write that a ratio ofapproximately 5 is beginning to be reasonable. Not all are of one mind on thisissue, however, Carmines and McIver (1981) suggesting the upper limit ofacceptability is three. The second measure used was that devised by Jöreskog andSörbom (1984), a goodness of fit index ranging between zero and unity, the latterindicating a perfect fit. The third was the root mean square error ofapproximation (RMSEA), a measure of model adequacy, compensated for modelcomplexity, based on the population discrepancy function. Browne and Cudeck(1993) suggest that a value of .08 or less for the RMSEA would indicate areasonable error of approximation but would not want to employ a model whoseRMSEA was greater than .10.

Team roles were progressively paired according to an established model ortheory; a non-systematic approach to reducing the number of traits by a randompairing of team roles was not followed. Thus, initially, team roles wereprogressively paired in accordance to Belbin’s suggestion, noted earlier, but theapproach failed to produce an acceptable fit between the correlated uniquenessmodel and the data sets. Similarly, pairing up team roles along the lines of “task”and “relationship” team roles proposed in Fisher et al. (1998) also failed toproduce an acceptable fit of the correlated uniqueness model to the data sets.However, in the work of Broucek and Randell (1996) on the construct validity ofthe Belbin team roles we noted they had correlated Belbin’s ObserverAssessment measure with the five-factor model of personality (Costa & McCrae,1992; Digman, 1990). Following their report we categorized a selection of theBelbin team roles into pairs so as to reflect the NEO-PI-R five factors model. TheNEO-PI-R five factors personality factors are shown in Table 4; they are:Extraversion , Conscientiousness, Openness, Agreeableness, and Neuroticism.Assigning four of the eight Belbin team roles to these five categories wasrelatively straightforward: the Resource investigator clearly aligned with theExtraversion factor, the Plant with the Openness, the Team worker with theAgreeableness , and the Shaper with the Neuroticism factor. These team roles aresharply defined and tend not to overlap with other team roles. However,categorizing the remaining four roles was harder. For them, Belbin’s descriptionsare broad. It is not clear, for example, if Chairmen are most notable for theiragreeableness or for their extraversion or, indeed, for their conscientiousness.Similarly, it was hard to decide if the Completer-finishers and Monitor-evaluators are more conscientious than neurotic, or vice versa. As for theCompany workers, they seemed to be able to fit in four out of the five categorieswithout difficulty. Some of the ways in which the eight Belbin team roles were


TA

BLE

4M

easu

res

of f

it in

dic

es fo

r th

e m

od

els

test

ed

Roo

t mea

nM

inim

umG

oodn

ess

squa

re e

rror

Mod

eldi

scre

panc

yof

ofno

.E

xtra

vers

ion

Con

scie

ntio

usne

ssO

penn

ess

Agr

eeab

lene

ssN

euro

tici

smfu

ncti

on/d

ffi

t ind

exap

prox

imat

ion

1ri

cw +

ch

pltw

cf +

me

+ s

h7.

12.8

8.1

36

2ri

ch+

cw+

me

pltw

cf+

sh6.

53.8

9.1

29

3ch

+ri

cf+

cw+

me

pltw

sh6.

18.8

9.1

25

4ri

cwpl

ch+

twcf

+m

e+sh

5.17

.91

.112

5ri

cf+

cwpl

ch+

twm

e+sh

5.31

.91

.114

6ri

cf+

cw+

me

plch

+tw

sh4.

89.9

2.1

09

7ri

cf+

me

plch

+cw

+tw

sh4.

90.9

2.1

09

8ch

+ri

me

plcw

+tw

cf+

sh4.

84.9

2.1

08

9ri

cw+

me

plch

+tw

cf+

sh4.

78.9

2.1

07

10ri

me

plch

+cw

+tw

cf+

sh4.

71.9

2.1

06

11ch

+ri

me

plcw

+tw

cf+

sh4.

49.9

2.1

03

12cw

+ri

me

plch

+tw

cf+

sh3.

81.9

3.0

92

cf=

Com

plet

er-f

inis

her;

ch=

Cha

irm

an;

cw=

Com

pany

wor

ker;

me=

Mon

itor

-eva

luat

or;

pl=

Pla

nt;

ri=

Res

ourc

e in

vest

igat

or; s

h=S

hape

r; tw

=T

eam

wor

ker;

df=

degr

ees

of f

reed

om (

50).

Tea

m r

ole

grou

ping

s

138


divided up between these five categories are shown in Table 4. Somecombinations gave data sets to which the correlated uniqueness model fittedpoorly, other less poorly, and a few reasonably well. The best combination foundwas model 12, shown in Table 4.

Validity and method effects

Convergent validity, discriminant validity, and method effects may be assessed,as they are in traditional confirmatory factor analysis (Kenny & Kashy, 1992)from the estimated parameters shown in Table 5. Inspection of the data, Table 5,obtained from fitting the correlated uniqueness model to the five trait team,model 12: (cw+ri), me, pl, (ch+tw), (cf+sh), reveals that:

(1) each of the factor loadings on each of the team types, that is on (cw+ri),me, pl, (ch+tw), and (cf+sh), is statistically significant. On this basisconvergent validity may be claimed for each of the five team types:(cw+ri), me, pl, (ch+tw), and (cf+sh)

(2) the means and deviations of the factor loadings on all three methods,16PF, OPQ, and video, are all sufficiently close to suggest that no onemethod contains more trait variance than the others

(3) only two out of the ten factor covariances are small and statisticallyinsignificant; this is a clear, strong signal indicating lack of discriminantvalidity

(4) numerous unique covariances for each of the methods, and particularly thepersonality questionnaire methods, are significant, thus indicating thepresence of method effects. Although the video technique is the best of thethree, it still shows the presence of method effect.

DISCUSSION

Application of the criteria of Campbell and Fiske (1959) to the multitrait–multimethod matrix revealed, as noted earlier, that the Belbin team role modelpossessed convergent validity but failed to exhibit discriminant validity; theevidence for the latter was much less than that for the former. At the morerigorous test of confirmatory factor analysis Belbin’s eight team role modelfailed totally, thereby disallowing any firmer conclusions concerning constructvalidities to be drawn beyond those reached with the Campbell and Fiske criteria.However, the confirmatory factor analysis result was a positive outcome in that itpointed to the problem of overparameterization, namely, that the data appearednot to be able to support eight distinct constructs. Thus, in an indirect manner theconfirmatory factor analysis result has further undermined the claim fordiscriminant validity for the eight role team model.

The failure of a theory to possess demonstrated and adequate psychometricproperties does, however, not necessarily prejudice its widespread acceptance


TA

BLE

5E

stim

ated

par

amet

ers

for

the

com

bin

ed B

elb

in te

am r

ole

s: C

orr

elat

ed u

niq

uen

ess

mo

del

Fac

tor

load

ings

Uni

que

vari

ance

and

cov

aria

nce

Var

iabl

e16

1718

1920

12

34

56

78

910

1112

1314

15

116

PF

(cw

+ri

)1.

000

00

02.

39*

216

PF (m

e)0

1.00

00

0–.

50*

2.34

*3

16P

F (p

l)0

01.

000

0–.

91*

.59*

2.50

*4

16P

F (

ch+

tw)

00

01.

000

2.27

*–.

66*

1.23

*2.

81*

516

PF

(cf+

sh)

00

00

1.00

–.38

1.06

*–.

07.0

92.

97*

6O

PQ

(cw

+ri

)1.

04*

00

00

00

00

04.

40*

7O

P’Q

(me)

0.9

3*0

00

00

00

01.

56*

2.85

*8

OP

Q (p

l)0

0.9

2*0

00

00

00

1.47

*1.

06*

2.72

*9

OP

Q (c

h+tw

)0

00

.71*

00

00

00

2.28

*.5

5*1.

30*

5.03

*10

OP

Q (c

f+sh

)0

00

0.5

5*0

00

00

3.51

*1.

24*

1.21

*.3

45.

76*

11V

ID (

cw+

ri)

.77*

00

00

00

00

00

00

00

4.60

*12

VID

(me)

0.9

1*0

00

00

00

00

00

00

.47*

2.63

*13

VID

(pl)

00

1.26

*0

00

00

00

00

00

0–.

24.4

11.

76*

14V

ID (

ch+

tw)

00

0.6

7*0

00

00

00

00

00

.83*

.02

.25

4.09

*15

VID

(cf

+sh

)0

00

0.5

8*0

00

00

00

00

01.

23*

1.14

*.2

5.5

9*5.

70*

Fac

tor

vari

ance

s an

d co

vari

ance

s16

cw+

ri3.

95*

17m

e–1

.29*

1.66

*18

pl–.

81*

.12

1.37

*19

ch+

tw3.

14*–

1.52

*–1.

19*

5.58

*20

cf+

sh–1

.72*

1.07

*.3

4–1

.99*

8.08

*

N=

331;

*p

< .0

5.

140


and perceived utility. A prime example is Schutz’s (1958/1989) FIRO-B theoryof interpersonal relations. The examination of the construct validity of FIRO-Bby Ryan, McGuire, and Ryan (1970) was the first in a series of investigations(Macrosson, 2000) of the eight scales of FIRO-B; all investigations havedemonstrated the scales’ lack of orthogonality. Throughout the years theseresearches were being published FIRO-B continued to be widely used bypersonnel practitioners in industry and commerce (Macrosson, 2000). It wouldappear that the same is true with the Belbin team role model. Although questionshave remained unanswered for many years about its psychometric properties, theBelbin team role theory’s use continued unabated (Broucek & Randell, 1996).

A successful resolution to the problem of the nonpositive definite estimatedcorrelation matrix was found in creating “condensed” versions of the Belbineight team role model. The ability to fit these “condensed” team roles satis-factorily to a five-trait correlated uniqueness model was confirmation that thedata could not support Belbin’s eight team role constructs. The best fitting ofthese “condensed” team role models, the five team role model 12 (see Figure 4),was shown to possess convergent, but to lack discriminant, validity, a result thatechoes that for the Belbin eight team role model. The lack of discriminantvalidity became obvious while fitting the correlated uniqueness model to the fiveteam role models. It was discovered at that point that the four Belbin team roles ofChairman, Company worker, Completer-finisher, and Monitor-evaluator couldbe readily interchanged without greatly altering the goodness of fit of the model.The large and significant factor covariances shown in Table 5 were, thus, only tobe expected. The definitions of some of the eight Belbin team roles appear to lieat the root of the problem. Two examples illustrate the problem: (1) theChairman’s behaviour of clarifying goals and getting things organized is hard todistinguish from the efficient Company worker who turns ideas into practicalaction, and (2) the Completer-finisher’s searching out of errors and omissions ishard to distinguish from the accurate judging and “seeing all options” of theMonitor-evaluator. Close reading of Belbin’s (1981) original exposition revealsmany such examples. Broucek and Randell (1996, p. 403) claimed that the Belbin“team role theory is flawed by conceptual discrepancies …”. It is hard to avoidconcluding that a core “conceptual discrepancy” is the overlapping team roledefinitions , which result in the observed poor discriminant validity.

Discriminant validity is assumed to be a good and desirable quality and, inregard to team role theories, can be seen as dependent on well-defined roles.However, McCrimmon (1995), on the basis of many years of consultancy workin industry and commerce, has taken a contrary view. He asserted that indi-viduals’ knowledge of their own roles coupled with a knowledge of team roletheory leads not only to an expectation that they will behave in accordance withtheir role at all times, but will also lead to a potentially restrictive obligation to doso. Additionally, he asserted there will be a reluctance to contribute from thepoint of view of someone else’s role lest they feel usurped. In the face of these


dangers of territoriality and abdication, McCrimmon (1995) proposed that teammembers allow themselves to move into other roles, subject to criterion of raisingteam effectiveness. His experience of teams, particularly in circumstances ofswiftly changing demands, permitted him to underscore the importance of flexi-bility. Plainly, operational plasticity and ease of movement between team roles isgreatly facilitated when team role definitions are ambiguous rather than clear.Lack of discriminant validity may confer benefit in the application of team roletheory.

Broucek and Randell (1996) suggested that the Belbin team roles may havemore than an intuitive appeal. They did not indicate what might lie beyond thatintuitive appeal, but in their correlation study with Belbin’s ObserverAssessment instrument and Costa and McCrae’s (1992) NEO-PI/NEO-PI-R FiveFactor Inventory, it is possible that they sensed that Belbin had, unwittingly andimprecisely , been identifying the “Big Five” while observing his teams engagedin their business simulation exercises. Indeed, the fit of the data to the five-rolecorrelated uniqueness model is sufficiently good as to constitute an invitation toadvance a “Big Five” team role model. A “Big Five” team role model may beseen to rest comfortably with the progression of tasks in the delivery of a projectas noted earlier, project tasks being: Planning; Defining resources needed;Scheduling; Implementing; Measuring progress; Reporting; and Maintaininghuman relations. Thus, “Openness” in the Five Factor Inventory subsumesinventive, insightful, enthusiastic, versatile, and imaginative behaviour, clearlyvirtues in the “Planning” and “Defining resources” phases; “Extraversion”subsumes outgoing, forceful and energetic behaviour, important inputs to the“Implementing” phase; “Conscientiousness” subsumes efficient, methodical andprecise behaviour, vital contributions to the “Scheduling” phase; “Neuroticism”subsumes anxious and fearful behaviour, key aspects for “Measuring progress”and “Reporting” phases; and “Agreeableness” subsumes trusting, friendly,stable, tolerant behaviour, indispensable traits for the “Maintaining humanrelations”. Additionally, fitting a “Big Five” categorization to a decision-makingprocess, for example, the general purpose consensus problem-solving model ofHurley (1995), is equally easy.

In conclusion, it is hard to avoid assigning Belbin’s team role theory to thecategory occupied by many commercially available products such as FIRO-Band the Myers-Briggs Personality Type Indicator, extensively employed butpsychometrically unsound. The possibility opened up by the findings reported inthis paper is that it may well be possible to create a team role theory based upona five-factor model of personality that has both psychometric validity as well asoperational utility. To such an end Barrick and Mount’s (1991) work has laid asecure foundation for a “Big Five” team role theory, since, in an extensive meta-analysis, they established that the “Big Five” are, indeed, valid predictors of jobperformance. Certainly, should a “Big Five” team role model ever be developed,it would have a more obvious grounding in theory than Belbin’s.


REFERENCES

Barrick, M.R., & Mount, M.K. (1991). The Big Five personality dimensions and job performance:A meta-analysis. Personnel Psychology , 44, 1–26.

Belbin, R.M. (1981). Management teams: Why they succeed or fail. London: Heinemann.Belbin, R.M. (1993). Team roles at work. London: Butterworth-Heinemann.Belbin, R.M., Aston, R.R. & Mottram, R.D. (1976). Building effective management teams.

Journal of General Management , 3(3), 23–29.Benne, K.D., & Sheats, P. (1948). Functional roles of group members. Journal of Social Issues, 4,

41–49.Berry, J.R. (1995). Team structure and management . Unpublished PhD thesis, Brunel University,

UK.Broucek, W.G., & Randell, G. (1996). An assessment of the construct validity of the Belbin Self-

Perception Inventory and Observer’s Assessment from the perspective of the five factor model.Journal of Occupational and Organizational Psychology , 68, 81–99.

Browne, M.W. (1982). Covariance structures. In D.M. Hawkins (Ed.), Topics in appliedmultivariate analysis (pp. 72–141). Cambridge, UK: Cambridge University Press.

Browne, M.W. (1984). Asymptotically distribution-free methods for the analysis of covariancestructures. British Journal of Mathematical and Statistical Psychology , 37, 62–83.

Browne, M.W., & Cudeck, R. (1993). Alternative ways of assessing model fit. In K.A. Bollen &J.S. Long (Eds.), Testing structural equation models. Newbury Park, CA: Sage Publications.

Campbell, D.T., & Fiske, D.W. (1959). Convergent and discriminant validity by the multitrait–multimethod matrix. Psychological Bulletin, 55(2), 81–105.

Carmines, E.G., & McIver, J.P. (1981). Analysing models with unobserved variables. In G.W.Bohrnstedt & E.F. Borgatta (Eds.), Social measurement: Current issues. Beverly Hills, CA:Sage Publications.

Cattell, R.B., Eber, H.W., & Tatsuoka, M.M. (1970). Handbook for the 16PF. Champaign, IL:Institute for Personality and Ability Testing.

Costa, P.T., & McCrae, R.R. (1992). Revised NEO Personality Inventory (NEO PI-R) and NEOFive-Factor Inventory (NEO-FFI) Professional Manual. Odessa, FL: PsychologicalAssessment Resources.

Digman, J.M.(1990). Personality structure: Emergence of the five-factor model. Annual Review ofPsychology , 41, 416–440.

Dulewicz, V. (1995). A validation of Belbin’s team roles from 16PF and OPQ using bosses’ ratingsof competence. Journal of Occupational and Organizational Psychology , 68, 81–99.

Fisher, S.G., Hunter, T.A., & Macrosson, W.D.K. (1998). The structure of Belbin’s team roles.Journal of Organizational and Occupational Psychology , 71, 283–288.

Fisher, S.G., Hunter, T.A., & Macrosson, W.D.K. (2000). The distribution of Belbin team rolesamong UK managers. Personnel Review, 29, 124–140.

Fisher, S.G., Macrosson, W.D.K., & Sharp, G. (1996). Further evidence concerning the Belbinteam-role self-perception inventory. Personnel Review, 25, 61–67.

Furnham, A., Steele, H., & Pendleton, D. (1993). A psychometric assessment of the Belbin team-role self-perception inventory. Journal of Occupational and Organizational Psychology , 66,245–257.

Gerth, H.H., & Mills, C.W. (1958). From Max Weber: Essays in sociology . New York: OxfordUniversity Press.

Gupta, S.K., & Taube, L.R. (1985). A state of the art survey of research on project management(1976–1983). In B.V. Dean (Ed.), Project management: Methods and studies. Amsterdam:Elsevier Science Publishers.

Huczynski, A. (1997). How to build successful teams … the Belbin way. ACCOUNT, 9 (3), 32–34.Hurley, W.B. (1995). A general purpose consensus problem-solving model. Team Performance

Management: An International Journal , 1, 26–35.


Jöreskog, K.G. (1974). Analysis of covariance structures. Scandinavian Journal of Statistics, 8,65–92.

Jöreskog, K.G., & Sörbom, D. (1981). LISREL V: Analysis of linear structural relations by themethod of maximum likelihood. Chicago: International Educational Services.

Jöreskog, K.G., & Sörbom, D. (1984). LISREL-7: User’s reference guide . Moorsville, IN:Scientific Software.

Kaiser, H.F. (1981). A revised measure of sampling adequacy for Factor Analytic Data Matrices.Educational and Psychological Measurement, 41, 379–381.

Kavanagh, M.J., MacKinney, A.C., & Wolins, L. (1971). Issues in managerial performance:Multitrait–multimethod analyses of ratings. Psychological Bulletin, 75(1), 34–49.

Kenny, D.A. (1979) Correlation and causality. New York: Wiley.Kenny, D.A., & Kashy, D.A. (1992). Analysis of the multitrait–multimethod matrix by

confirmatory factor analysis. Psychological Bulletin, 112, 165–172.Life, E.A. (1993). Calculating team role scores from 16PF Primary Factor Sten Scores . Windsor,

UK: NFER-NELSON.Mabey, W., & Hunter, R. (1986). Using personality measures to improve selection. Guidance and

Assessment Review, 2, 1–4.Macrosson, W.D.K. (2000). FIRO-B: Factors and facets. Psychological Reports, 86, 311–320.Marsh, H.W. (1989). Multitrait–multimethod analysis. In J.P. Keeves (Ed.), Educational research

methodology, measurement and evaluation: An international handbook . Oxford, UK:Pergamon Press.

Marsh, H.W. (1990). Confirmatory factor analysis of multitrait–multimethod data: The constructvalidation of multidimesional self-concept responses. Journal of Personality, 58(4), 661–692.

Marsh, H.W., & Hocevar, D. (1983). Confirmatory factor analysis of multitrait–multimethodmatrices. Journal of Educational Measurement , 20(3), 231–248.

Marsh, H.W., & Hocevar, D. (1985). Application of confirmatory factor analysis to the study ofself concept: First- and higher-order factor models and their invariance across groups.Psychological Bulletin, 97, 562–582.

McCrimmon (1995). Teams without roles: Empowering teams for greater creativity. Journal ofManagement Development , 14(6), 35–41.

Roderick, C. (1997a). Team roles: The dark side (Part 1). Selection and Development Review,13(5).

Roderick, C. (1997b). Team roles: The dark side (Part 2). Selection and Development Review,13(6), 18–19.

Ryan, B.A., Maguire, T.O., & Ryan, T.M. (1970). An examination of the construct validity of theFIRO-B. Journal of Projective Techniques and Personality Assessment, 34, 419–425.

Saville, P., Holdsworth, R., Nyfield, G., Cramp L., & Mabey, W. (1992) Occupational PersonalityQuestionnaire Manual. Esher, Surrey: Saville & Holdsworth Ltd.

Schutz, W. (1989). FIRO-B: A three-dimensional theory of interpersonal behaviour. Mill Valley,CA: Will Schutz Associates. (Original work published 1958). [Also reprinted as Schutz, W.(1978). The interpersonal underworld. Palo Alto, CA: Consulting Psychologists].

Senior, B. (1997). Team roles and team performance: Is there “really” a link? Journal ofOrganizational and Occupational Psychology , 70, 241–258.

Wheaton, B., Muthén, B., Alwin, D.F., & Summers, G.F. (1977). Assessing reliability and stabilityin panel models. In D.R. Heise (Ed.), Sociological methodology 1977 (pp. 84–136). SanFrancisco: Jossey-Bass.

Widaman, K.F. (1985). Hierarchically nested covariance structure models for multitrait–multimethod data. Applied Psychological Measurement, 9, 1–26.

Wothke, W. (1993). Nonpositive definite matrices in structural modelling. In K.A. Bollen &J.S. Long (Eds.), Testing structural equations models. Newbury Park, CA: Sage.

Manuscript received May 2000Revised manuscript received January 2001

A Validation Study of BELBINs TEAM ROLES

Documents