A Theory-Based Measure of Conflict Management Strategies in the Workplace Author(s): Carsten K. W.

De Dreu, Arne Evers, Bianca Beersma, Esther S. Kluwer, Aukje Nauta Reviewed work(s): Source: Journal of Organizational Behavior, Vol. 22, No. 6 (Sep., 2001), pp. 645-668 Published by: John Wiley & Sons Stable URL: http://www.jstor.org/stable/3649613 . Accessed: 19/02/2012 23:00
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

John Wiley & Sons is collaborating with JSTOR to digitize, preserve and extend access to Journal of Organizational Behavior.

http://www.jstor.org

Journal of OrganizationalBehavior J. Organiz.Behav. 22, 645-668 (2001) DOI: 10.1002/job.107

A theory-based measure of conflict management strategies in the workplace

CARSTEN K. W. DE DREUI*, ARNE EVERS', BIANCA BEERSMA' ESTHER S. KLUWER2 AND AUKJE NAUTA3
Amsterdam,The Netherlands 1Universityof Amsterdam, 2 Utrecht University,Utrecht,The Netherlands 3 Universityof Groningen,Groningen,The Netherlands

Summary

Conflict managementinfluences individualwellbeing, group performanceand organizational effectiveness.This researchexaminedthe psychometricqualitiesof two versions of the newly developed test for conflict handling.The lean version (Study 1 and 2) included problem solving, forcing, yielding and avoiding as distinct conflict management strategies, and the expanded version (Study 3) also included compromising. A negotiation study (Study 1) and observer rated showed substantialconvergence between self-reports,opponent-reports behaviorfor problemsolving, forcing andyielding, but not for avoiding.In Study 2 and Study 3 the psychometricpropertieswere examined of the lean and the expandedversion, respecfactoranalyses revealedgood to excellent psychometricqualities of both tively. Confirmatory flexible andvalid instrument versionsof the scale. We concludethatthe scale is a parsimonious, to assess conflict managementstrategiesat work. Copyright? 2001 JohnWiley & Sons, Ltd.

Introduction
The effectiveness of individual employees, teams and entire organizations depends on how they manage interpersonal conflict at work (Tjosvold, 1998). Managers spend an average of 20 per cent of their time managing conflict (Thomas, 1992), and evidence suggests conflict and conflict management at work substantially influences individual, group and organizational effectiveness, as well as wellbeing, as indicated by health complaints and doctor visits (De Dreu et al., 1999; Spector and Jex, 1998). Given the importance of conflict management in organizations, it is vital to have and develop reliable and valid measurement instruments. Such instruments help researchers to obtain valid data through self and peer-report, and practitioners to diagnose conflict management strategies at work. Unfortunately, the measurement instruments that have been described in the literature either suffer from low psychometric quality, or the psychometric quality is unknown. The present research was designed

* of to: 15, Correspondence CarstenK.W. De Dreu, Universityof Amsterdam,Department Psychology, Roetersstraat 1018 WB Amsterdam,The Netherlands. E-mail: ao_dedreu@macmail.psy.uva.nl

Copyright? 2001 John Wiley & Sons, Ltd.

Received 25 February2000 Accepted 9 April 2001

646

C. K. W.DE DREUETAL.

to assess the psychometricqualities of a revised and updatedversion of the Dutch Test for Conflict Handling(DUTCH) designed by Vande Vliert (1997). The test has been developedby Dutch scholars but is not necessarily applicable only to people from the Dutch culture. As we will elaborate upon below, the theoretical basis for the test generalizes across culture and the original test corresponds closely with tests developed in the United States (Van de Vliert, 1997). As an outline, we firstdiscuss the theoreticalbasis underlyingthe instrument, we discuss several and conflict managementtests that have been describedin the literatureand precededthe DUTCH. Subsequently,we reportthreestudiesdesignedto assess the psychometric qualitiesof the DUTCH.In Studies 1 and2 we examineda 'lean' versionof the DUTCH which measuresproblemsolving, forcing,yielding and avoiding.In Study3 we examinedan 'expanded'versionwhich includescompromisingas a distinct strategy.We concludewith a discussionof some practicaland theoreticalimplicationsof the results.

Conflict management at work: dual concern theory

Conflictmanagementis what people who experienceconflict intendto do as well as what they actually do (Vande Vliert, 1997). Although an infinite numberof conflict managementstrategiesmay be conceived of, conflict researchand theory tends to converge on Dual ConcernTheory (Pruittand Rubin, 1986). Dual ConcernTheory is relatedto earlierwork by Blake and Mouton (1964) and to Deutsch's Theory of Cooperationand Competition(Deutsch, 1973). It arguesthat conflict managementis a function of high or low concernfor self, combined with high or low concernfor others.As FigureI shows, high concern for self and low concernfor othersresults in a preferenceforforcing, focused on imposand ing one's will on others.Forcinginvolves threatsandbluffs, persuasivearguments, positionalcommitments.Low concernfor self and high concernfor othersresults in a preferencefor yielding, which is orientedtowardsaccepting and incorporating others will. It involves unilateralconcessions, unconditionalpromises, and offering help. Low concern for self and othersresults in a preferencefor avoiding, which involves reducingthe importanceof the issues, and attemptsto suppressthinkingaboutthe issues. High concern for self and others producesa preferencefor problemsolving, which is oriented towardsan agreementthatsatisfiesboth own andothers' aspirationsas much as possible. It involves an

High

Yielding

Problem Solving

Concern For Other

(Compromising)

Low

Avoiding Low

Forcing High

for Concern Self

for of as of strategies a function concern self Figure1. Theoretical management representation the fiveconflict andconcern other for
( Copyright 2001 JohnWiley& Sons,Ltd.

Behav. 645-668 (2001) 22, J. Organiz.

CONFLICT MANAGEMENT 647

exchange of informationabout priorities and preferences, showing insights, and making trade-offs between importantand unimportant issues. some authorshave suggested that intermediateconcern for self, paired to intermediate Recently, concern for othersresults in a preferencefor compromising.Some see compromisingas 'half-hearted problem solving (e.g., Pruitt and Rubin, 1986). Others, however, see it as a distinct strategy that involves the matchingof others' concessions, makingconditionalpromises and threats,and an active searchfor a middle ground(e.g., Van de Vliert, 1997). Ratherthan seeking to settle this debate at the theoreticallevel, we choose to test two versions of the DUTCH. The 'lean' version includes problem solving, forcing, yielding and avoiding (Studies 1 and 2); the 'expanded'version (Study 3) includes compromisingas well. Doing so yields informationabout the psychometricpropertiesof two related measurementscales. In addition, Study 3 will provide some empirical argumentsfor or against the conceptualizationof compromisingas a distinct conflict managementstrategy.

Conflict management tests

Dual Concern Theory has received good support in field research in organizations (Blake and Mouton, 1964; Thomas, 1992; Van de Vliert, 1997) as well as in experimentalresearchin the social psychological laboratory(Carnevale and Pruitt, 1992; De Dreu et al., 2000). As such, the theory provides a solid basis for the development of instruments to assess conflict management strategies at work. Several such instruments exist. One of the oldest methods is the Conflict Measurement Survey (e.g., Kilmann and Thomas, 1977), but it has been criticized and replaced by other measures because of disappointing psychometric qualities (Landy, 1978; Thomas and Kilmann, 1978; Shockley-Zalabak, 1988). Substantial improvement was reached by the development of the Rahim OrganizationalConflict Inventory (ROCI-2). The instrumentcomes in three forms, one concerned with conflicts with one's supervisor (Form A), one concerned with conflicts with one's peer (Form B), and one concerned with conflicts with one's subordinates (Form C). In each case the instrumentmeasures the five conflict management strategies discussed above. Rahim and Magner(1995) conductedconfirmatory analysis of data from one studentand four professional samples, and showedconvergentanddiscriminant validity of the five sub-scales of measuring conflict management.Close inspectionof the goodness-of-fit(GFI) indices in both studentand professional samplessuggests moderatefit falling below the standard cut-off criterionof 0.90. Satisfactoryfit was reachedonly when within each sub-scaleitems were aggregatedinto two sets (RahimandMagner, 1995, p. 126). Although this procedureis statistically and methodologically correct, it makes the instrumentunfit for applied purposes in which one wishes to quickly and unequivocally answer the question which conflict managementstrategy respondentsuse within a particularsetting. Thus, the ROCI-2is widely used but it suffers from a numberof disadvantagesincluding less than optimal psychometricproperties. To redressthese shortcomings,Van de Vliert and colleagues developed a test for conflict handling (EuwemaandVande Vliert, 1990; JanssenandVande Vliert, 1996; Vande Vliert, 1997). In interactive sessions with focus groups, new items were designed and items derived from the ROCI-2 and other instrumentswere revised and adaptedup to a point where the coefficient alphas and patternsof intercorrelationswere sufficient to good (Van de Vliert, personal communication,June 14, 1999). Comparedto the ROCI-2,the DUTCH has fewer items (16 or 20 instead of 28), which is importantwhen we want to measureconflict managementas partof a large organizationalsurvey and need as short a measure as possible. Also, the DUTCH leaves the hierarchicalrelationshipbetween conflict parties unspecified,which increases the investigator'sflexibility. Copyright 2001 JohnWiley& Sons,Ltd. ? 22, J. Organiz. Behav. 645-668 (2001)

648

C. K. W.DE DREUETAL.

The problem solving (forcing) scale of the DUTCH is positively (negatively) related to integrative agreements in group negotiation (Beersma and De Dreu, 1999). However, we have no information about the convergence between self and other reports of conflict management strategies, or between self-reports and actual conflict behavior. Study 1 was designed to fill this void and examined the convergentvalidity of the DUTCH by comparingself-ratingsof forcing, problem solving, yielding and avoiding during a negotiation task with ratings by one's opponent and with observer ratings of conflict behavior. Also, Study 1 assessed the extent to which selfreports are susceptible to self-serving bias - the tendency to see one's own conflict management as more constructive and less destructive than the conflict management of one's opponent (De Dreu et al., 1995). In addition,the psychometricqualities of the DUTCH are unknown.To fill this void, we assessed in Study 2 the psychometric qualities of the lean version of the DUTCH. Study 3 was designed to assess the psychometricqualities of the expanded version of the DUTCH (including compromising as a distinct strategy). In Studies 2 and 3, we focused on (1) the unidimensionalityof the four or five sub-scales, (2) the discriminantvalidity, and (3) the invarianceacross gender and hierarchical level.

STUDY 1 Method
Participantsand procedure
Seventy eight psychology students(29 males and 49 females) at the Universityof Groningenparticipated in this study.They were randomlyassigned to dyads. The mean age of the participantswas 22 years, ranging from 19 to 26 years. Initial analyses revealed no differences between male and female participants. Participantscame to the laboratoryin groups of four and care was taken that participantsin one session were unacquainted. in ourpast research(De Dreu et al., 1998, 1995 - Study 3), participants As were seatedin separaterooms andhad 10 minutesto readtheirrole instructions(see below). They were to imagine that they were in the describedsituation,and to preparea conversationwith anotherparticipant. Hereafter,participantswere paired and seated in separaterooms where they engaged in an audiotapednegotiation(see below). After 15 minutes,the experimenterenteredthe room, and handed were debriefedand dismissed. out a questionnaire. participants Upon completionof the questionnaire,

Negotiation task
The study employed a role playing methodology that was developed in a pilot study and was based on our past research (e.g., De Dreu et al., 1998, 1995 - Study 3). Participantswere provided with backgroundinformationin which a conflict situation was described and then freely interactedwith each other to discuss the issue. The instructionsstated that the participantworks togetherwith a fellow student on a researchproject. The original schedule of activities for the next week is that both studentsdo statisticalanalyses togetherbut that this original schedule is no longer applicableand that Copyright 2001 JohnWiley& Sons,Ltd. ? Behav. 645-668 (2001) J. Organiz. 22,

CONFLICT MANAGEMENT 649

they need to reach an agreement about how to change the schedule of activities. Thus, members within one dyad needed to reach an agreement about who would do what kind of activity the next week. Through instructions,parties were given opposed preferences for certain activities and they were given argumentsthey could use to defend their particularpreferences. Participantswere told, finally,that they had 15 minutes to discuss their opposing preferencesand to reach an agreementon a new schedule of activities. (In contrastto some of our past research,participantswere not given a pay-off schedule indicating the value of a particularagreement.Rather,they were told what agreements they preferredand what agreementsthey did not prefer. As a result, the study did not yield objective performancedata).

Dependentvariables
Self-report and opponent-report measures In the post-negotiationquestionnaire,own conflict managementwas measuredwith the DUTCH. The instrumentcomprised 16 items, with four (randomlyordered)items per conflict managementstrategy (forcing, problemsolving, yielding, and avoiding; see Appendix A). Questionscould be answeredon 5-point scales, with 1 = not at all, to 5 = very much). Cronbach'salphaswere 0.86 for forcing, 0.81 to for problemsolving, 0.71 for yielding, and 0.69 for avoiding. In addition,we asked participants fill out the DUTCHwith regardto theiropponent'sconflict management.Cronbach'salphaswere 0.83 for forcing, 0.82 for problemsolving, 0.73 for yielding, and0.67 for avoiding.The orderin which own and others' conflict managementwas assessed was variedsystematically.Because it had no effects it is not discussed any further. Observer measures The negotiationswere audiotapedand subsequentlycoded by two trained observersunawareof the goals of the study. The trained observers independentlyrated the conflict behaviors of each of the two partieswithin a dyad. Using electronic beeps, the 38 audiotapedconversationswere divided into time samples of two minutes (n = 212 time samples, M = 5.4 time samples per dyad). Every twominute interval,the observersrated four conflict behaviors (problem solving, forcing, yielding, and reliavoiding; 1 = demonstratednot at all, to 4 = demonstratedto a great extent). Inter-observer abilities were sufficientto good, with Cronbach'salphas rangingfrom 0.61 to 0.81 for forcing. Two reliabilities for yielding (a = 0.53) and for avoiding (a = 0.32). An exceptions were the interrater is that both constructs had low variance (SD = 0.49 and SD = 0.16 respectively). explanation When looking at absolute reliability, it appearedthat 141 out of 212 times (67per cent) observers agreed upon yielding and that 200 out of 212 times (94 per cent) observers agreed upon avoiding reliabilities for the four scales were sufficient to (Cohen's Ks > 0.67). We conclude that interrater good.

Results
Withineach dyad we had observerratings,and self and otherreportsfor each party(furtherreferredto as PartyA and PartyB) on each of the four sub-scales from the DUTCH. Since partiesinteractedtheir datawere interdependent we decided to analyse the data for each partyindependentlyand to treat and datafor PartyB as a replicationof datafor PartyA. This is a conservativeapproachbecause it reduces Copyright ) 2001 JohnWiley& Sons,Ltd. Behav. 645-668 (2001) J. Organiz. 22,

650

C. K. W. DE DREU ETAL.

our degrees of freedomby half. It is, however,justified in light of the fact that in the negotiationParty A and PartyB were given differentpositions with differentarguments.

Convergentand divergentvalidity
The upperpartof Table 1a gives the means, standard between obserdeviations,and inter-correlations ver ratingsof PartyA and opponent-ratings PartyA on the one hand, and self-reportsby PartyA on of the other.The lower part of Table la gives the same informationwith regardto PartyB. Before discussing observed correlations,we wish to emphasize that moderatelynegative or moderatelypositive correlationsmay be expected, theoretically,between differentsub-scales because they both load high on one dimension (e.g., concern for self) and low on anotherdimension (e.g., concern for self). For example, problem solving may be positively correlatedwith yielding (both share a high concern for other),while forcing and yielding may be expected to correlatenegatively(they differ both on concern for self and concernfor other).In otherwords, we expected strongpositive correlationson the diagonal in Table la, and sometimes moderatelynegative, or moderatelypositive correlationsoff the diagonal in Table la. As can be seen in Table la, self-reportedforcing was positively and stronglycorrelatedwith observer ratingsof forcing and with opponent-ratings forcing. Moreover,self-reportedforcing was negaof related with the other conflict managementstrategiesratedby observersor the opponent.This tively suggests that self-reportedforcing has high convergentand divergentvalidity.A similarbut somewhat weakerpatternemergedfor problemsolving. Again, self-reported problemsolving had strongpositive correlationswith observerratings and opponentratings of problem solving, and positive but weaker correlationswith yielding. Interestingly,problemsolving and avoiding were positively, but not significantly, correlated.This may be understoodin terms of Dual ConcernTheory,where problemsolving is the result of high dual concern, and avoiding the result of low dual concern. Thus, problemsolving tends to have good convergentand acceptable divergentvalidity. Withregardto self-reportedyielding, dataare supportiveof convergentvalidity,in thatself-reported yielding is positively correlatedwith observerratingsand opponentratingsof yielding behavior.One exception to this is that self-reportedyielding correlatedas positive with observerratingsfor yielding as with observerratingsfor problemsolving, suggesting thatobservershave difficultyseparatingmere concession making from problem solving behavior.This can be understoodwhen we realize that in negotiation,problemsolving also involves logrolling - making small concessions on importantissues and large concessions on unimportant issues. Table 1a shows, finally,thatavoidinghad poor convergentvalidity.Both for PartyA and for PartyB data, self-reportedavoidingdid not correlatewith observerratingsand with opponentratingsof avoiding. Most likely, the lack of convergentand divergentvalidity is due to the fact that avoidingbehavior is difficultto observe from audiotapedconversation.We returnto this in the Conclusion and General Discussion section. and Besides correlationsbetween self-reportand opponent-report observerreport,it may be useful to examine correlationsbetween the sub-scales for self-reports.Table lb provides these. As can be seen, the patternof intercorrelations closely resembles those reportedin Table 1a. For both PartyA and B, problem solving and yielding, problem solving and avoiding, and yielding and avoiding were not related. For both Party A and Party B, forcing and problem solving were negatively and significantly related, and forcing and yielding were negatively but not significantly related. Somewhat troublesome is the strong negative correlationbetween forcing and avoiding for the Party A data, which were not replicatedin the PartyB data, and was not presentin Table 1a either.This may reflect measurementproblems related to the avoiding sub-scale. Copyright ) 2001 JohnWiley& Sons,Ltd. Behav. 645-668 (2001) J. Organiz. 22,

00 ?s

t"
t'

Table la. Means, standarddeviations for self-reportedconflict management,and correlationsbetween self-report, oberver conflict behaviors;party A (top panel) and Party B (bottom panel) M
00

SD 1.36 1.24 0.97 0.96 1.41 1.43 1.05 1.07

Forcing(Ob) Forcing(Op) ProbSol(Ob) ProbSol(Op) Yield(Ob) 0.72* - 0.45* - 0.03 - 0.40* 0.41* -0.18 - 0.43* 0.01 0.33t - 0.36t - 0.41* -0.11 0.31t -0.43* - 0.46* -0.23 - 0.55* 0.45* 0.36t 0.33t - 0.38t 0.42* 0.46* -0.17 - 0.45* 0.47* 0.20 0.22 - 0.22 0.54* 0.35t -0.06 - 0.38t 0.17 0.38t 0.25 - 0.31 0.17 0.39t -0.25

Yie -

.,e

2 tt

Forcing Problem Solving Yielding Avoiding Forcing Problem Solving Yielding Avoiding

3.28 3.22 2.97 1.72 2.90 3.55 2.97 1.91

Note:*p< 0.01;t p < 0.05 (two-tailed, with n = 38; (Ob) = observerratings;(Op) = opponentratings.

652

C. K. W. DE DREUETAL.

and A for Tablelb. Correlations conflictmanagement strategies Party (abovethe diagonal) amongself-report B Party (belowthe diagonal) Forcing
Forcing Problem solving Yielding - 0.54* - 0.32

Problem solving
- 0.47* 0.35

Yielding
- 0.23 0.29 -

Avoiding
- 0.61* 0.04 0.29

Avoiding
< Note.*p 0.001, withn = 38.

0.08

-0.16

0.13

Table2. Meansfor conflictmanagement a function target judgment as of of Conflict management Forcing of Target Judgment
Self 3.09*t

Problem solving
3.38*

Yielding
2.97t

Avoiding
1.81l

2.91t 2.98t t-tests 37). to means donotshare same that the differ (df= superscript atp < 0.01accordingpaired Note:*,t,l Other 3.12*t

1.82'

Self-serving bias in self-reportsof conflict managementstrategies

To assess self-servingbias in self-reportedconflict managementwe conductedan Analysis of Variance with Target of Judgment (self or opponent) and Conflict Management (forcing, problem solving, variables.Results revealedsignificantmain effects for Target yielding, avoiding) as within-participant of Judgment,F(1, 37) = 66.11, p < 0.001, and for Conflict Management,F(3, 35) = 11.50, p < 0.001, as well as a significantinteractionbetween Targetof Judgmentand ConflictManagement,F(3, 35) = 12.77, p < 0.001. Cell means are given in Table 2. Pairedt-tests revealedno effects for Target of Judgment(self versus opponent)on forcing, yielding or avoiding, all ts (37) < 1, n.s. For problem solving, however, self-ratingswere higher than opponentratings,t (37) = 3.60, p < 0.001. Thus, there is evidence for self-serving bias for problem solving, but not for forcing, yielding and avoiding. It shouldbe noted thatevidence for self-servingbias in the perceptionof problemsolving not necessarily means lack of convergentvalidity between the self-reportand opponentreportmeasures.Convergent validity is assessed on the basis of correlations,which were high and positive (cf., Table la), while self-serving bias is assessed by comparingmeans (cf., Table 2). What we can conclude is that and the opponentconverge in their conclusion that self engages in more proalthoughthe participant blem solving, participants rate the amountof own problem solving as higher than does the opponent.

STUDY 2
Study 1 showed acceptable convergence between self, opponent and observerratings of forcing and problemsolving, to some extent for ratingsof yielding, but not for ratingsof avoiding. Self-reportsfor yielding and avoiding had moderateto poor convergentand divergentvalidity and this may be due to the fact that these sub-scales of the DUTCH have poor psychometricproperties.Study 2 was conducted to assess the psychometricqualities of the DUTCH. We focused on (1) the unidimensionality of the four sub-scales, and (2) the discriminantvalidity. 0 Copyright 2001 JohnWiley& Sons,Ltd. Behav.22, 645-668 (2001) J. Organiz.

CONFLICT MANAGEMENT 653

ContextualSidebar

Organizational Context for Study 2

Organization The site concerns one of the most profitablebusiness units of a large manufacturing organization producingheavy machinery.At the time of the study (1999), the manufacturing organization'sstock value is substantiallyundervalued the possibility for take-overis a continuingissue duringthe and time of the study.Threeyearsbeforethe studywas conducted,a largerestructuring had operation been a transition fromindividual-based team-basedpracticesalong with downto implemented,involving sizing and reductionof the numberof managementstaff. Employees Employees in the firmareunionized.The union contractpermitteda limited numberof sick days for which the employee received full pay - unlimited additionalsick days (up to two years) could be taken at 70 per cent salary, provided a doctor's certificatewas provided. The restructuring programme had resulted in a number of employees working in new positions that were considered less interestingand providinglower status. This was particularlytrue for a small numberof maintenancestaff and offshoreengineerswho, because of theirfrequentabsencewere not includedin the study. Morale Employees traditionallyderiveda strong sense of pride working for the company,which provided them with a strongidentity andrelativelygood workingconditions,pay and benefitspackages.The economic changes of the nineties, makingold industryless interestingto work for, underminedthis sense of pride and the restructuring programmeinduced some unrest and uncertaintiesas well as role ambiguitiesand sense of increasedwork pressures.

Method
Sample
of were employeesof a companyspecializedin the developmentandconstruction food proRespondents On the initiativeof both managementand laborunions a projectwith regardto work cessing systems. stress was started.The DUTCH was includedin the instrument assess the level of stress and several to relatedconcepts.All employees (includingemployees from the generalmanagementand financialand sales departments, well as employees in productionor technicalpositions) were asked to participate. as Out of 364 employees 308 returned questionnaire the (responserate = 85percent). Seven questionnaires were not included in the analyses because of one or more missings. Mean age and length of of appointment the respondentswere 43.91 (SD = 10.41) and 21.32 (SD - 13.01) years,respectively. Eleven per cent of the respondentswere female. 0 Copyright 2001 JohnWiley& Sons,Ltd. Behav. 645-668 (2001) 22, J. Organiz.

654

C. K. W.DE DREUETAL.

Procedure
The DUTCH was partof a booklet which containedin all about250 items. The other items dealt with sources of stress, health, work satisfaction, control, etc. The items of the DUTCH were in random order.A researchassistantinvited employees to participatein the study, and an accompanyingcover letter from the firm's CEO emphasized the importanceof participatingas well as its voluntaryand anonymous character.The cover letter furtherexplained the purpose of the study, which was to get a better understandingof the way employees work together, experience their work environment (e.g., technical equipment,internalcommunications)to be able to improvethe situationwhere necessary. The researchassistantadministeredthe questionnairesin small groups (it was sent to the home were filled addressof employees who were not able to attendthese groupsessions). The questionnaires in anonymouslyand collected by the researchassistant.

Analysis and Results

and interrelationsof thefour sub-scales Unidimensionality
The internalconsistency reliabilitycoefficients (coefficient a) were computedfor the four scales. The minimal acceptablelevel of alphawas set to be 0.60. (Fortests thatareprimarilymeantto studygroups a reliability coefficient of between 0.60 and 0.70 qualifies as 'sufficient' and a coefficient of 0.70 or higher as 'good' [Bartram,1997; Evers, et al., 2000]). Since a high alpha is no guaranteefor unidimensionality (Crockerand Algina, 1986), LISREL analyses were conductedto check for unidimensionality.By means of this analysis we checked for unwantedhigh correlationsbetween pairsof items that were too similar in content, that is items that cause an 'unfair' increase in alpha. We used the fit indices of X2 (divided by df), the adjusted goodness-of-fit index (AGFI) and root mean square of approximation(RMSEA), which are importantfor scale development (Bentler, 1990; Schumacker and Lomax, 1996). A low x2 (when judged against degrees of freedom) is suggestive of good fit. An AGFI above 0.90 is indicative of good fit and above 0.95 of excellent fit. The RMSEA residuals should be about 0.05 or smaller for a good fit. Table 3 reportsthe basic psychometricpropertiesof the DUTCH scales. The internalconsistencies can be qualifiedas sufficient,and, except for two RMSEA figureswhich are somewhattoo high, all fit indices show good fit. The AGFI values reveal excellent fit. Apparentlythe items fit well into their respective scales and the high fit indices and low RMSEA values show that no unwanted overlap between items within a scale is detected. From these datawe conclude thateach scale separatelydelivers reliable informationwith respect to unidimensionallymeasuredconstructs.

Table 3. Internalconsistencies and LISRELfit indices of the DUTCH scales

a
Yielding Dominating Collaborating Avoiding 0.65 0.83 0.73 0.64

x2
1.95 5.57 1.32 5.76

df
2 2 2 2

AGFI
0.99 0.96 0.99 0.95

RMSEA
0.00 0.07 0.00 0.08

Copyright 0 2001 John Wiley & Sons, Ltd.

J. Organiz.Behav. 22, 645-668 (2001)

CONFLICT MANAGEMENT 655 Table 4. Goodness-of-fitindices for one-, two-, three-, and four-factormodels Model Null model One-factormodela Two-factormodelb Three-factormodelc Four-factormodeld 2 1316.05* 836.78* 552.14* 356.42* 233.08* df 120 104 103 101 98 AGFI 0.54 0.59 0.72 0.81 0.87 RMSEA 0.21 0.18 0.13 0.10 0.07

= Note.AGFI adjusted error approximation. of RMSEA= rootmeansquare index; goodness-of-fit *p< 0.001. aDifference one-factor nullmodel: and X2(16)= 479.27.* = bDifference andone-factor two model: X2(1) 284.64.*
cDifferencethree and two-factormodel: X2(2) = 195.72.* dDifference andthree-factor four model: X2(3)= 123.34.*

The next step was to test the proposed four-factormodel by means of a maximum likelihood LISREL analysis of the covariancematrix of the 16 DUTCH items. In this model each of the items was allowed to load on its associated factor only, and the four factors (representingthe four scales) were not allowed to correlate.The same fit indices as mentionedabove were used. Following Rahim and Magner (1995), we also computed the GFI for the independentor null-model and one- through four-factormodels. The two-factormodel grouped(a) problemsolving and yielding; and (b) forcing and avoiding (cf. Deutsch, 1973). In the three-factormodel only the items of the sub-scales avoiding and yielding were groupedtogether,whereas the remainingtwo factors comprisedthe same items as in the four-factormodel (cf., Putnam and Wilson, 1982). Table 4 summarizesthe measures of the GFI for the null- and one- throughfour-factormodels. All measures show a gradual increase, and is the four-factormodel is superiorto all other models. The X2/df-ratio good; however, the AGFI is a little bit too low and the RMSEA is a little bit too high. Nevertheless the values are indicative of a sufficientto good fit. In all, the resultsshow thatthe four-factormodel constitutesa good representation of the interrelationsamong the 16 items of the DUTCH.

Discriminantvalidity
The third step in the analyses was to assess the discriminantvalidity of the scales of the DUTCH by computing the correlationsbetween the latent factors. The values on the diagonal were set to 1 (the completely standardizedsolution). Table 5 shows the factor intercorrelationsfor the total between group.The results supportthe discriminantvalidity of the scales, in that the intercorrelations scales are generally low (the correlationof 0.33 between yielding and avoiding being the highest correlation).

Table5. Factor intercorrelation matrix = 301) (n DUTCH-factors 2. Dominating

1. Yielding

1 0.12 (0.07)
0.12 (0.11) 0.33 (0.23)

3. Collaborating 4. Avoiding

0.00 (- 0.02) 0.13 (0.14)

- 0.23 (- 0.08)

Note:Indices brackets correlations in are manifest variables. between Copyright(0 2001 John Wiley & Sons, Ltd. J. Organiz.Behav. 22, 645-668 (2001)

656

C. K. W.DE DREUETAL.

STUDY 3
Study 1 and 2 revealedthatthe lean version of the DUTCH has acceptableto good psychometricproperties. As mentioned at the outset, however, some scholars arguethat in additionto the four strategies identified in Dual Concern Theory, compromisingshould be considered as a distinct strategyrather than as a 'lazy form of problem solving' (cf. Pruitt and Rubin, 1986; Van de Vliert, 1997). If true, we should be able to measure compromising,and confirmatoryfactor analyses should reveal better fit for a five-factorratherthan a four-factormodel. In addition,multidimensionalscaling shouldreveal a patternof inter-correlations fits a two-dimensionalspace with forcing, problemsolving, yielding that and avoiding occupying cornerpositions in a 2 x 2 space and compromisingoccupying a centreposition (cf., Figure 1). To examine these issues, Study 3 examined the psychometricpropertiesof a fivescale version of the DUTCH. The instrumentis identical to the one used in Study 1 and 2 but also includes four items to measure compromising.

Method
Sample
Data were collected from a heterogeneous sample of professionals (n = 2403). Mean age and work experienceof the respondentswere 34.14 (SD = 8.41) and 5.52 (SD= 5.73) years, respectively.Thirty per cent of the respondentswere female.

Data gathering procedure

The DUTCH was placed on the web-site of the Dutch weekly magazine Intermediair(publishedby Elsevier/ Reed). The magazine is targetedtowardshigher educatedprofessionalsin the fields of manand agement,consulting,personnelcounselling, financialadministration controlling,and law. Its aims include providing information about career possibilities and updates on technological and human resource developmentsin various professional fields. The web-site has been installed to supportthe printedpaper version of the magazine, and contains various kinds of information,including several tests designed to screenone's own tendenciesandpolicies. The DUTCH was partof this series of tests. Respondentswere asked to answerseveralbackgroundquestionstappinginto theirdemographiccharacteristics and their work situation. Subsequently, they were presented with the 20 items of the DUTCH (four items measuring forcing, problem solving, compromising, yielding, and avoiding; see AppendixA). Questionswere presentedin a randomorder.In between December 1998 andMarch 1999, 2403 respondentscompleted the questionnaire.We screened the answers(includingbiographical data) to eliminate respondentsthat filled out the questionnairemore than once. This reduced the sample size to 2182. Measures The backgroundquestions assessed gender and age. In addition, respondentswere asked whether they experienced most conflicts in their work situationwith (1) their superior(s), (2) their peers and Copyright 2001 JohnWiley& Sons,Ltd. ? Behav. 645-668 (2001) J. Organiz. 22,

CONFLICT MANAGEMENT 657 or colleagues, (3) their subordinate(s), (4) other.Subsequently, respondentswere asked 'how they tend to deal with conflict in their daily work situations.'The 20 items of the DUTCH were presentedin a randomorder.When respondentshad answeredall questions,they were providedwith a shortdescription of the five conflict managementstrategies(respondentscould not enterthis site when they had not answeredall questions). This descriptionalso contained some references for furtherreading, in case respondentswere interestedin learningmore aboutconflict and conflict managementin organizations.

Analysis and Results

and interrelationsof thefive sub-scales Unidimensionality
Table6 reportsthe basic psychometricpropertiesof the DUTCH sub-scales.The internalconsistencies can be qualifiedas sufficient, and except for one X2/df-ratio, fit indices show good fit. The AGFI all values reveal excellent fit. Apparentlythe items fit well into their respective scales and no unwanted overlapbetween items within a scale is detected. From these data we conclude that each scale separately delivers reliable informationwith respect to unidimensionallymeasuredconstructs. The next step was to test the proposedfive-factormodel by means of a maximumlikelihood LISREL analysis of the covariancematrixof the 20 DUTCH items. In this model each of the items was the allowed to load on its associatedfactoronly, and the five factors(representing five scales) were not allowed to correlate.In additionto the models computedin Study 2 we computeda five-factormodel which included compromisingas a distinct strategy.Table 7 summarizesthe measuresof the GFI for Table6. Internal consistencies LISREL indicesof the DUTCH scales and fit a Yielding Compromising Forcing Problem solving Avoiding
*p < 0.01; tp < 0.001.

X2 0.44 8.49 1.73 19.50t 9.25*

df 2 2 2 2 2

AGFI 1.00 0.99 1.00 0.98 0.99

RMSEA 0.00 0.04 0.00 0.06 0.04

0.65 0.66 0.70 0.68 0.73

models Table7. Goodness-of-fit indicesfor one-,two-,three-, four-,andfive-factor Model

Null model One-factormodela Two-factormodelb Three-factormodelc Four-factormodeld Five-factormodele
*p < 0.001.

X2
9513.84** 5798.63** 4358.11** 2752.42** 1980.36** 1214.77**

df
190 170 169 167 164 160

AGFI
0.55 0.64 0.74 0.83 0.87 0.93

RMSEA
0.19 0.16 0.12 0.10 0.08 0.05

Note. AGFI= AdjustedGoodness-of-FitIndex; RMSEA = Root Mean SquareErrorof Approximation. aDifferenceone-factorand null model: X2(20)= 3815.21.* bDifferencetwo and one-factormodel: X2(1)= 1440.52.* cDifferencethree and two-factormodel: X2(2)= 1605.69.*
dDifference four and three-factor model:

eDifferencefour and five-factormodel: X'(4) = 765.59.*

2(3)= 772.06.*

Copyright( 2001 John Wiley & Sons, Ltd.

J. Organiz.Behav. 22, 645-668 (2001)

658

C. K. W. DE DREU ETAL.

the null- and one- throughfive-factormodels. All measuresshow a gradualincrease,andthe five-factor model is superiorto all other models. The AGFI and RMSEAvalues indicate a good fit. Although the X2/df-ratio too high, our very large sample size can be held responsible for this high value of X2, is since X2is dependenton sample size. A reanalysiswith a randomsample of 1000 respondentsresulted in a decreaseof x2 to 556, whereasthe otherindices remainedthe same. This shows thatthe five-factor model constitutes a good representation the interrelationsamong the 20 items of the DUTCH. of

Discriminantvalidity
The thirdstep in the analyses was to assess the discriminantvalidity of the scales of the DUTCH by computing the correlationsbetween the latent factors. The values on the diagonal were set to 1 (the for completely standardizedsolution). Tables 8 and 9 show the factor intercorrelations the total and gendergroups,and for the groupsthatdiffer in hierarchicallevel of the conflict opponentrespectively. Over all sub-samplesthe correlationsshow a consistentpattern.Althoughthereare some fluctuations, these rarelyexceed 0.10. The results supportthe discriminantvalidityof the scales, as the intercorrelations between scales are generallylow (the correlationsbetween yielding and compromising,yielding and avoiding,and compromisingandproblemsolving are moderate,but for each scale enough specific varianceremains).

Invarianceacross gender and hierarchicallevel

The fourthstep in the analyses was to examine the invarianceof the five-factormodel of the DUTCH across different groups. Two analyses were performed:one for gender, and one for the hierarchical level of the conflict opponent.For each analysis the covariancematricesfor the 20 DUTCH items were computed.We verifiedwhetherage was relatedto conflict management,and resultsshowed it was not.
Table 8. Factor Intercorrelationmatrixfor total and gender groups

DUTCH-factors
Totalgroup (n = 2182) 1. Yielding 2. Compromising 3. Forcing 4. Problem solving 5. Avoiding Men (n = 1543) 1. Yielding 2. Compromising 3. Forcing 4. Problem solving 5. Avoiding Women(n = 639) 1. Yielding 2. Compromising 3. Forcing 4. Problem solving 5. Avoiding

0.41 -0.28 0.19 0.51

(0.20) ( - 0.15) (0.10) (0.39)

0.02 (0.04) 0.50 (0.33) 0.22 (0.20)

0.06 (0.08) - 0.21 (- 0.15)

- 0.20 (-0.11)

0.37 -0.26 0.19 0.51

(0.23) (-0.16) (0.11) (0.39)

0.08 (0.08) 0.52 (0.35) 0.20 (0.19)

0.12 (0.11) - 0.21 (- 0.16)

- 0.18 (-0.10)

0.50 (0.30) - 0.35 ( - 0.20) 0.20 (0.07) 0.51 (0.39)

- 0.12 (- 0.04) 0.46 (0.30) 0.27 (0.25)

- 0.06 (0.00) - 0.21 (- 0.14)

- 0.24 (- 0.13)

Note:Indices brackets correlations in are between manifest variables. Copyright( 2001 John Wiley & Sons, Ltd. J. Organiz.Behav. 22, 645-668 (2001)

CONFLICT MANAGEMENT 659 Table9. Factor intercorrelation matrix hierarchical for relation groups DUTCH-factors 1. Yielding 2. Compromising 3. Forcing 4. Problem solving 5. Avoiding
Conflictwith superior (n = 564)

0.46 (0.31) - 0.17 (- 0.09) 0.22 (0.12) 0.50 (0.38) 0.40 (0.23)

0.15 (0.14) 0.43 (0.29) 0.24 (0.21)

0.06 (0.08) - 0.18 (-0.13)

- 0.17 (-0.10)

1. Yielding 2. Compromising 4. Problem solving

3. Forcing

Conflictwith colleague (n = 736)

-

- 0.41 ( - 0.22)

5. Avoiding 0.62 (0.43) with subordinate(n = 310) Conflict

0.03 (0.01)

- 0.07 (0.00)

0.50 (0.34)
0.22 (0.20)

0.07 (0.08)
- 0.22 ( - 0.16)

- 0.28 ( - 0.16)

1. Yielding 2. Compromising 3. Forcing 4. Problem solving 5. Avoiding

0.51 (0.33) - 0.40 ( - 0.27) 0.33 (0.18) 0.47 (0.38)

- 0.16 ( - 0.08) 0.54 (0.36) - 0.09 (-0.02) 0.28 (0.25) -0.19 (-0.14)

-0.05 (-0.01)

in Note: Indices brackets correlations are between manifest variables. For the total sample, the correlationwith age was r = 0.06, - 0.03, - 0.08, 0.06, and 0.05 for avoiding, dominating,compromising,yielding and problemsolving, respectively.Furtheranalyses revealed no differences between male and female respondents. We estimatedsix models using LISRELand comparedthem sequentially.In Model 1 the patternof factor loadings is held invariant.In Model 2 the patternof factor loadings, and the factor loadings are held invariant. Model 3 the patternof factorloadings, the factorloadings, and the item interceptsare In held invariant.In Model 4 the patternof factor loadings, the factor loadings, the item intercepts,and the errorsare held invariant.In Model 5 the patternof factor loadings, the factor loadings, the item are intercepts,the errors,and the variances/covariances held invariant.In Model 6, finally,the pattern and of factorloadings,the factorloadings,the item intercepts,the errors,the variances/covariances, the mean are held invariant.Besides 2/ df, and the RMSEA for these analyses the GFI- and Akaike's informationcriterion (AIC) indices are reported.The interpretationof the GFI index is similar to AGFI. AIC is a badness-of-fitindicator,with small values indicating good fits and large values poor ones. It is addedbecause the differencein AIC when imposing more restrictionsgives additionalinforwill only be a mationaboutthe parsimonyof the model. DecreasingX2 solving for more parameters by is the case AIC will benefit if X2 is decreased by more than 2.0 for each parameteradded. If this decrease as well (Loehlin, 1998). Tables 10 and 11 present the results with respect to the analysis of the factor invariance.Table 10 shows the results for gender,and Table 11 for the hierarchicallevel of the conflict opponent. (Participants were allowed to indicate whether they primarily had conflicts with their superior(s),their colleague(s), their subordinate(s),or 'other.'Those classifying themselves in the lattercategory are left out of these analyses.) Note that the firstthreemodels in this analysis are the most critical. The difference in x2 between Model 2 and Model 1 was significant for gender,indicatingthatthe hypothesisof equal factorloadings for men andwomen mustbe rejected. However, the magnitude of the differences in Table 10 suggests this is not a severe problem. The tests for the differences of all other consecutive models were negative, implying equal item and intercepts,errors,variances/covariances, means for men and women. As AIC graduallydecreases (except for the step from Model 1 to Model 2), one can conclude that the more restrictions are Copyright 2001 JohnWiley& Sons,Ltd. ? Behav. 645-668 (2001) J. Organiz. 22,

660

C. K. W.DE DREUETAL.

Table 10. Invarianceanalysis across gender Model 1. Equal factorpattern 2. Model 1 + equal factor loadingst 3. Model 2 + equal item interceptst 4. Model 3 + equal errors? 5. Model 4 + equal variances/covariances1 6. Model 5 + equal means: error approximation. of *p< 0.001. TModel2 - Model1: X2(15)= 32.56(p < 0.01).
tModel 3 - model 2: X2(15)= 9.46 (p <0.85). "Model4 - model 3: X2(20) = 27.22 (p < 0.13). 5 SModel - model4: X2(15)= 19.31(p<0.20). CModel6 - model 5: X2( 5) = 6.67 (p < 0.25).

AIC 1751.39 1752.72 1731.14 1722.97 1710.95 1706.79

GFI 0.92 0.92 0.92 0.92 0.92 0.92

X2 1394.94* 1427.50* 1436.96* 1464.18* 1483.49* 1490.16*

df 320 335 350 370 385 390

RMSEA 0.06 0.06 0.06 0.05 0.05 0.05

Note. Male = 1543; female = 639. AIC = Akaike's informationcriterion;GFI= goodness-of-fit;RMSEA = root mean square

Table 11. Invarianceanalysis across hierarchicalrelationshipto conflict opponent Model AIC

E2 GFI

df 1315.16* 1346.40* 1371.69* 1414.52* 1467.59* 1476.44* 480 510 540 580 610 620

RMSEA 0.06 0.06 0.05 0.05 0.05 0.05

1. Equal factor pattern 1757.22 2. Model 1 + equal factor loadings 1721.85 3. Model 2 + equal item interceptst 1686.84 4. Model 3 + equal errors" 1644.84 5. Model 4 + equal variances/covariances'1647.97 6. Model 5 + equal means 1638.65
Note. Conflict with superior =564; Model 2 IModel 3 "Model4 4Model 5 9Model 6

0.91 0.90 0.90 0.90 0.89 0.89

= information of error approximation. RMSEA rootmeansquare criterion; = goodness-of-fit; GFI

= conclict with colleague =736; conflict with subordinate 310. AIC = Akaike's

*p<0.001.
-

Model 1: X2(30)= 31.26 (p < 0.40). Model 2: X2(30)= 25.29 (p < 0.71). Model 3: X2(40) = 42.83 (p < 0.35). Model 4: X2(30) = 53.07 (p < 0.01). Model 5: X2(10) = 8.85 (p < 0.55).

imposed upon the data the better they fit. This provides strong evidence of factor invariance for the inventoryacross gender. The analyses for the variousgroupsthatare formedwith respectto the hierarchicallevel of the conflict opponent show the hypotheses of equal factor loadings, equal item intercepts,equal errors,and equal means cannotbe rejectedfor these groups (see Table 11). However,the differencein x2 between Model 4 andModel 5 was significant,indicatingthatthe hypothesisof equal variancesandcovariances for must be rejected.For all the otherrestrictionsthe model is invariant hierarchicallevel, and with all the restrictionsimposed the model seems to fit best (as shown by the decrease in AIC). Thus, we can conclude that the analyses supportthe factor invarianceof the inventoryfor hierarchicallevel of the conflict opponent as well.

Multidimensionalscaling
Although the results show that the DUTCH tends to have good psychometric quality, the question remains whether the ordering of the five sub-scales correspondsto what one would predict on the
Copyright ? 2001 John Wiley & Sons, Ltd. J. Organiz.Behav. 22, 645-668 (2001)

CONFLICT MANAGEMENT 661

1.5

Problem Solving
o

0.5

Yielding

o Concern forOther

Compromising

-0.5

Avoiding Forcing
-1.5

-2

-1.5

-1

-0.5

0.5

1.5

Concern Self for of as of Figure2. Empirical representation the five conflictmanagement strategies a function two dimensions for for (concern self andconcern other) basis of the Dual Concern Theory (see Figure 1). To answer this question, we conducted a multidimensional scaling of the data. The purpose of this procedureis to find a configurationof points whose Euclideanoutputdistances reflect as closely as possible the rankorderof inputdissimilarities. Assuming the data fit one dimension (e.g., competition-cooperation) yielded an unsatisfactoryfit (Young's S = 0.19, with R2 = 0.87). Assuming two dimensions (i.e., concern for one's own outcomes, and concern for other) results revealed that Young's stress index was 0.00251 (0.00147 for the matrix), with R2 = 0.99, which is both very good. The solution is displayed in Figure 2. As can be seen, compromisingtends to be in the middle, while forcing, avoiding, problem solving and yielding tend towardsthe cornersof the 2 x 2 matrix.It shouldbe mentioned,however,that yielding is too close to avoiding. Nevertheless, the patternof (Euclidean) distances tends to corroboratethe theoreticalpatterndepicted in Figure 1.

Conclusionsand GeneralDiscussion
To study conflict managementstrategiesin the workplace,reliable and valid instrumentsare crucial. instrumentsthat survey in the literatureand are used by scientists and practitionersto Unfortunately, assess (preferencesfor) conflict managementstrategiesmay be criticized on methodologicalor practical grounds.The currentresearchbuilds upon earlier work by Van de Vliert (1997) and was concerned with the psychometric qualities in the Dutch Test for Conflict Handling (DUTCH). We soughtanswersto two questions.First,do self-reportsof conflict managementstrategiesconvergewith ( Copyright 2001 JohnWiley& Sons,Ltd. Behav. 645-668 (2001) 22, J. Organiz.

662

C. K. W. DE DREU ETAL.

what behavior? conflict withactual and Second, other(i.e., opponent) reports do self-reports converge In of arethepsychometric two of theDUTCH? examined versions theDUTCH. Studies1 We qualities In and and2 we useda 16-item instrument measures that yielding, avoiding. solving,forcing, problem to in addition the otherfour 3 we examined 20-iteminstrument measured a that Study compromising conflictmanagement strategies.

Theoreticalimplications
on and of between solving,forcing, yielding theone Study1 showed convergence self-reports problem the revealed expected andopponent-reports thesestrategies the other. of on Moreover, hand, Study1 behaof correlations between of and ratings conflict solvingandobserver self-reports forcing problem the psychofor vior duringnegotiation. Less supportive resultswere obtained avoiding. Although of metricqualitiesof the sub-scale avoidingweregood, self-reports avoidingdid not converge for is A withreports theconflict or ratings. possibleexplanation thatavoiding by opponent, withobserver a For is the moreambiguous attributions. example, conflictpartywhoconto multiple strategy open to the of sistently downplays importance theconflictissuemaydo thisin order avoidtheissueandto as suchbehavior a cunning reduceinteraction a minimum. opponent, to The however, mayperceive to get one'sway,to buytimeandto imposeone'swill on others avoiding, (i.e., forcing). Perhaps way to that involvesbehaviors are difficult morethanany of the otherconflictmanagement strategies, the Because individual moreimportant. intentions accurate of underlying judge,making understanding observers and has betterknowledge thanopponents neutral aboutone's own intentions convergence is betweenself-reports other,reports avoiding likelyto be low. and of is is literature whether An issuethathasgenerated somedebate theconflict in compromising a 'lazy or a trulydistinctconflictmanagement formof problem and strategy solving'(Pruitt Rubin,1986) if thanempirical. rather However, (Vande Vliert,1997).Up to now,the debatehasbeenconceptual from be ableto designitemsthataredistinct is one compromising indeeda distinctstrategy should fit reveala better should factoranalysis itemsdesigned measure to problem solvingandconfirmatory In can bothconditions be satisfied. addition, fora five-factor a four-factor than model.Study3 showed for concern self andinterif compromising a trulydistinctstrategy resultsfromintermediate that is that of reveala pattern inter-correlations mediate concern others, multidimensional for scalingshould this 3 (i.e., Figure showed in 2) Theory closelyfitstheDualConcern depicted Figure1. Again,Study in the andyieldingoccupied condition be satisfied: can positions solving avoiding, forcing,problem a tendedtowards midpoint fourcomersof a two-dimensional position. space,while compromising and of Thispattern resultsspeaksto theconstruct of measuring validity the instrument the sub-scale to these in Moreimportantly, resultscontribute conflicttheoryin thatthey compromising particular. fromproblem is distinct that for lendfurther solving support the argument compromising (empirical) thatfuture research, theory 1990;Vande Vliert,1997).Itis concluded (cf.,Vande VliertandKabanoff, as to wouldbenefit fromcloserattention compromising a distinct and strategy development, practice for managing conflictat work. for and, indirect evidence theDUTCH for Thepsychometric Theory support DualConcern provides
as such, complementsexperimentaland field research(for reviews, see De Dreu et al., 2000; Van de Vliert, 1997). Although Dual Concern Theory has been aroundfor quite some time, many scholars continue to work with related but differentmodels. For instance, the basic distinctionbetween competition and cooperationcontinues to inspire much theoreticaland empiricalwork (for a review, see Tjosvold, 1998). The currentresearchindicates this basic distinctionto be too broadand we suggest that future research and practice would benefit from the more fine-grainedanalysis offered by Dual ConcernTheory.
( Copyright 2001 JohnWiley& Sons,Ltd.

J. Organiz. Behav. 645-668 (2001) 22,

MANAGEMENT 663 CONFLICT The results of the three studies togethersuggest that for problemsolving, forcing and yielding (but not for avoiding) self-reports converge with opponent-reportsand with actual conflict behavior. has Results of Studies 2 and 3 show thatthe lean and expandedversion of the instrument good psychometric qualities. Convergentand discriminantvalidity were good, and invarianceacross gender and hierarchicalrelationshipwas high. These studies show that the DUTCH is a psychometricallysound instrument thathas, except for avoiding,good predictivevalidity.The DUTCH appearsto be a flexible and parsimoniousinstrumentthat may be of use to both researchersinterested in the relationships between conflict managementstrategiesand relatedaspects of organizational life, as well as to practitionersinterestedin diagnosing (preferencesfor) conflict managementstrategiesin organizational setdirect, empiricalcomparisonsawait futureresearch,the psychometricqualities of the tings. Although DUTCH appearto compare favorablyto other instrumentsin the literatureand future researchmay benefit from using the DUTCH.

Self-servingbiases
Study 1 revealed that self-ratingsfor problem solving were significantlyhigher than ratingsby one's conflict opponent.This may reflecta self-servingbias in the assessmentof one's conflict management. Self-serving bias has been observed in a wide varietyof settings including conflicts in organizational settings, in close relationshipsand in interpersonal negotiation(for a review, see Johns, 1998). Current results are consistent with this past research.Our results suggest social desirabilitydid not play an role because social desirabilitywould lead one to over-estimateone's tendencytowardsproimportant blem solving and to underestimate one's tendencyto engage in forcing. The latterwas not observed.It well be that people tend to underestimate extent to which their opponentengages in problem the may Conflict researchsuggests people tend to overly emphasize other's negative traits and behasolving. viors (De Dreu et al., 1995) and tend to engage in 'hostile attributions'- suspecting ulteriormotives underlyingother's ambiguousbehavior (Baron, 1997). An interestingavenue for futureresearchis to examine the conditions that foster or inhibit the individual's tendency to underestimate opponent'stendency to engage in problemsolving. The study the De Dreu et al. (1995) suggested that as conflict intensifies self-serving bias becomes more proby nounced,presumablybecause conflict partiesfeel more threatenedand develop a more negative view of their opponent(cf., Rubin et al., 1994). A recent study by Gelfand et al. (2000 - manuscriptunder review) showed thatnegotiatorsin an individualisticculture,such as the U.S. or the Netherlands,were more likely to develop self-serving perceptionsof own and other'sconflict managementthannegotiators in a collectivist culture,such as Japan.They explained this culturaldifference in termsof the collectivist negotiator's stronger inclination to define oneself in terms of interdependentrather than independentsocial structures. Together,these studies suggest that self-serving perceptionsof conflict managementdevelop especially when conflict partiesdefinethemselves as differentratherthan similar from their conflict opponent.

Limitationsand avenuesfor future research

The currentresearchhas strengthsas well as weaknesses. We examined the psychometricqualities of the DUTCH in a variety of samples (psychology studentsin Study 1; employees in a manufacturing firmin Study 2; and a large sample of higher educatedprofessionalsin Study 3). In addition,we used multiple ways to gather data, including paper-pencil and computerizedmethodology. Consistency across studies suggests the specific way in which data are gathered does not substantiallyimpact Copyright 2001 JohnWiley& Sons,Ltd. ? Behav. 645-668 (2001) J. Organiz. 22,

664

C. K. W. DE DREU ETAL.

the validity of the (sub) scales, but more controlledstudies are needed to settle this issue. Also, results in Study 3 may have been influencedby self-selection, in thatwe had no controlover who did, andwho did not visit the web site, and who did decide (not) to participate the study.Finally,we includedtwo in versions of the DUTCH, a lean version that includedproblemsolving, forcing, yielding and avoiding, and an expandedversion that also included compromising.Although both versions appearedto have sound psychometric qualities, the drawbackof this approachis that we have no data regardingthe convergence between self-reports and other-reportsfor compromising. This issue needs to be addressedin futureresearch. To understand results for avoiding we distinguishedbetween intentionsand (observable)behathe viors. Likewise, in discussing the self-serving perceptionsof conflict managementstrategies,we distinguished between intentions and behavior. In the conflict literature,this distinction is not always explicit. Although we asked people what they do when in conflict, it may be safe to assume that the DUTCH is a measureof the individual'sbehavioralintentions,ratherthan his or her actual behavior. Sometimes, intentionstranslateinto actualbehavior,but sometimes they do not. Sometimes,intentions do translateinto behaviorbut it is not perceived as such. Conflict theory,and futureresearchon scale development, may benefit from a more thoroughanalysis of what individualsin conflict want to do, what they believe they should do, and what they actually do. As mentionedat the outset,the DUTCHis not necessarilylimitedto the Dutchculture.The theoretical basis fortheinstrument and boundaries, we expectthepsychometric properties generalizesacrosscultural culturesrespondto fromdifferent to be constantacrossculture.This is notto say,however,thatindividuals conflictin similarways. Cultural differencesshouldrevealitself, however,in differencesin the meansand is Future research needed,hownot in differencesin factorstructures relatedpsychometric and properties. to test the psychometricpropertiesof the DUTCH in differentcultures. ever,

Preferencesfor conflict management:person or situation?

Before closing, we wish to addressthe questionof whetherconflict managementis stable over time, or is the primarily productof the situation.In Dual ConcernTheoryconflictmanagement seen as theproduct of concernfor self and concernfor others(see Figure 1). Concernfor self andconcernfor othersare, in and and turn,predicted one's personality the situation(De Dreuet al., 2000; Pruitt Rubin,1986;Vande by with concern correlate Forinstance,stableindividualdifferencesin social valueorientation Vliert, 1997). for others:Partieswith a pro-socialorientationhave higherconcernfor othersthanpartieswith an individualisticor competitivevalue orientation. Likewise,evidence suggests stableindividualdifferencesin collectivistversusindividualistic valuespredictconcernfor othersandthatconflictpartiesthatscorehigh on agreeablenessalso tend to have higherconcernfor others.At the same time, instructions constituby and the ents, incentivestructures, expectationof cooperativefutureinteraction, positivemood inductions all increaseconcernfor others(De Dreu et al., 2000). The fact that conflict managementis the productof both personalityand the situationdoes not suggest necessarily that within a particularwork setting conflict managementcannot be accuratelypredicted on the basis of instrumentslike the DUTCH. Worksettings tend to remainrelativelystable over do time. Employees interactwith the same co-workers,incentive structures not change overnightand employees do the same kind of work for longer periods of time and thus face the same (interpersonal) tend to basis. In addition,individualswithin the same unit, teamor department problemson a recurring influence one another(Salancik and Pfeffer, 1977), thus creating their own social environmentwith, most likely, ratherstable preferencesfor dealing with conflict. The consequenceis that an individual's actual and preferredconflict managementstrategies are likely to be relatively stable over time. We cannotknow, however,whetherthis is due to stable individualdifferences,to the fact thatthe situation Copyright 2001 JohnWiley& Sons,Ltd. ? Behav. 645-668 (2001) 22, J. Organiz.

MANAGEMENT 665 CONFLICT

is relativelyimperviousto change, or both. Consistentwith Dual ConcernTheory,however,we expect thatthe individual'sconflict managementstrategiesat work are relatively stable over time due to both stable individualdifferences and the relatively stable situationsin which people work.

Conclusion
The currentresearchreveals, firstof all, that the DUTCH is a valid and reliable instrumentthat can be used to measureconflict managementstrategiesin the workplace.The DUTCH may well complement the existing conflict managementtests especially when flexibility and parsimonyare importantand psychometric quality is valued. Second, the current research indicates that conflict theory would improve by incorporatingcompromising as a separateconflict managementstrategy that is distinct from problem solving. Third, this study indicates that future researchwould benefit from using the DUTCH to assess an individual'sconflict managementstrategies.Finally, this researchsuggests that practitionersshould no longer rely on a broadcompetition-cooperationdistinction,but consequently and consistently resortto a five-factortaxonomy of conflict managementstrategieswhen intervening in conflict situations,or assessing conflict managementstrategiesin a particularorganization.

Authorbiographies
Carsten K.W. De Dreu is Professorof Organizational Psychology at the Universityof Amsterdamand Director of Researchof the KurtLewin GraduateSchool for Social Psychology. His researchis concerned with social influence, group decision making, negotiation and organizationalconflict. Arne Evers is Associate Professorof Organizational Psychology at the Universityof Amsterdam.His researchis concernedwith test developmentand validation,occupationalstress and vocationalchoice. Bianca Beersma is a PhD studentin Organizational Psychology at the Universityof Amsterdam.Her researchconcerns the motivational,cognitive and structuralaspects of group negotiation, and team performance. Esther Kluwer is AssistantProfessorin Social and Organizational Psychology at UtrechtUniversity. Her researchinterests are social conflict and social justice in close relationshipsand gender issues. Aukje Nauta is Assistant Professor in the Department of Management and Organization at the University of Groningen, and Researcher at TNO Work and Employment. Her research interests include conflict and negotiation in organizations,organizationalcommitment, and human resource practices.

References
and a Anderson Gerbing 1988.Structural DW. in JC, approach. equation modelling practice: review recommended Van In BaronRA. 1997.Positiveeffectsof conflict: insightsfromsocialcognition. De DreuCKW, de VliertE Psychological SocietyBooks:Leicester.
Psychological Bulletin 103: 411-423. (eds). Using Conflictin Organizations,Sage: London; 177-191. Bartram D. 1997. Review of ability and aptitude tests (Level A) for use in occupational settings. British

( Copyright 2001 JohnWiley& Sons,Ltd.

Behav.22, 645-668 (2001) J. Organiz.

666

C. K. W. DE DREUETAL.

Beersma B, De Dreu CKW. 1999. Negotiation processes and outcomes in prosocially and egoistically motivated groups.InternationalJournal of ConflictManagement10: 385-402. Bentler PM. 1990. Comparativefit indexes in structural models. Psychological Bulletin 107: 238-246. Blake R, Mouton JS. 1964. The Managerial Grid. Gulf: Houston, TX. CarnevalePJD, PruittDG. 1992. Negotiation and mediation.AnnualReview of Psychology 43: 531-582. Crocker L, Algina J. 1986. Introductionto Classical and Modern Test Theory. Holt, Rinehart & Winston: Orlando,FL. De Dreu CKW,NautaA, Vande Vliert E. 1995. Self-serving evaluationof conflict behaviorand escalationof the dispute.Journal of Applied Social Psychology 25: 2049-2066. De Dreu CKW, Harinck F, Van Vianen AEM. 1999. Conflict and performancein groups and organizations. InternationalReview of Industrialand OrganizationalPsychology 14: 376-405. De DreuCKW,Giebels E, Vande Vliert E. 1998. Social motives andtrustin integrativenegotiation:the disruptive effects of punitive capability.Journal of Applied Psyhcology 83: 408-422. De Dreu CKW, WeingartLR, Kwon S. 2000. Influence of social motives on integrativenegotiation: a metaanalyticalreview and test of two theories.Journal of Personalityand Social Psychology 78: 889-905. Deutsch M. 1973. The resolutionof conflict: constructiveand destructiveprocesses. Yale UniversityPress: New Haven, CT. Euwema MC, Van de Vliert E. 1990. Gedragen escalatie bij hierarchischeconflicten (Behaviorand escalationin hierarchicalconflicts). ToegepasteSociale Psychologie 4: 28-41. Evers A, Frese M, Cooper CL. 2000. Revisions and furtherdevelopmentsof the OccupationalStress Indicator (OCI):LISRELresultsfrom four DUTCH studies.Journalof Occupationaland Organizational Psychology 73: 221-240. JanssenO, Vande Vliert E. 1996. Concernfor other'sgoals: key to de-escalationof conflict.InternationalJournal of ConflictManagement7: 99-120. Johns G. 1998. A multi-level theory of self-servingbehaviorin and by organizations.Researchin Organizational Behavior 21: 1-38. KillmanRH, Thomas KW. 1977. Developing a forced choice measureof conflict handlingbehavior.Educational and Psychological Measurement37: 309-325. Vol. 2, Landy F. 1978. Conflict managementsurvey. In Buros OK (ed). Eigth mental MeasurementYearbook, Gryphon:HighlandPark,N.J; 1173-1174. Loehlin JC. 1998. Latent VariableModels. An Introductionto Path, and StructuralAnalysis, 3rd ed. Factor, LawrenceErlbaum:Mahwah.NJ. PruittDG. 1998. Social conflict. In GilbertD, Fiske ST, Lindzey G. (eds). Handbookof Social Psychology,4th edn, Vol. 2, McGraw-Hill:New York; 89-150. PruittDG, Rubin J. 1986. Social Conflict:Escalation, Stalemateand Settlement.RandomHouse: New York. PutnamLL, Wilson CE. 1982. Communicativestrategiesin organizationalconflicts: reliability and validity of a measurementscale. In Communication Vol. 6, BurgoonM. (ed). Sage: Beverly Hills, CA; 629-652. Yearbook, Rahim A, Magner NR. 1995. Confirmatory factor analysis of the styles of handlinginterpersonalconflict: firstorder factor model and its invarianceacross groups. Journal of Applied Psychology 80: 122-132. Rubin JZ, Pruitt DG, Kim S. 1994. Social Conflict: Escalation, Stalemate, and Settlement. McGraw-Hill: New York. Salancik GR, Pfeffer J. 1977. An examination of need satisfaction models of job satisfaction.Administrative Science Quarterly22: 427-456. SchumackerRE, Lomax RG. 1996. A Beginners Guide to StructuralEquation Modelling. Lawrence Erlbaum: Mahwah, NJ. Shockley-ZalabakP. 1988. Assessing the Hall Conflict Management Survey. Management Communication Quarterly1: 302-320. Spector PE, Jex SM. 1998. Development of four self-reportmeasures of job stressorsand strain:interpersonal conflict at work scale, organizational workloadinventory,andphysical symptoms constraintsscale, quantitative inventory.Journal of OccupationalHealth Psychology 3: 356-367. Thomas KW. 1992. Conflict and negotiationprocesses in organizations.In Dunnette MD, Hough LM (eds). Handbook of Industrial and Organizational Psychology, 2nd ed, Vol. 3, ConsultingPsychologists Press: Palo Alto, CA; 651-717. Thomas KW, Killman RH. 1978. Comparisonof four instrumentsmeasuringconflict behavior.Psychological Reports 42: 1139-1145.

Copyright? 2001 John Wiley & Sons, Ltd.

J. Organiz.Behav. 22, 645-668 (2001)

CONFLICT MANAGEMENT 667 Tjosvold D. 1998. Cooperative and competitive goal approachto conflict: accomplishments and challenges. Applied Psychology: An InternationalReview 47: 285-342. Van de Vliert E. 1997. ComplexInterpersonalConflictBehavior Psychology Press: London. Van de Vliert E, Kabanoff B. 1990. Toward theory-based measures of conflict management. Academy of ManagementJournal 33: 199-209.

Copyright( 2001 John Wiley & Sons, Ltd.

J. Organiz.Behav. 22, 645-668 (2001)

668

C. K. W.DE DREUETAL.

Appendix A
The Dutch Testfor ConflictHandling (DUTCH) WhenI have a conflict at work, I do the following:
Yielding
1. 2. 3. 4. 5. 6. 7. 8. I give in to the wishes of the other party. I concur with the other party. I try to accommodatethe other party. I adaptto the other parties' goals and interests. I try to realize a middle-of-the-roadsolution. I emphasize that we have to find a compromise solution. I insist we both give in a little. I strive wheneverpossible towardsa fifty-fifty compromise. I push my own point of view. I search for gains. I fight for a good outcome for myself. I do everythingto win. I examine issues until I find a solution that really satisfies me and the other party. I stand for my own and other's goals and interests. I examine ideas from both sides to find a mutuallyoptimal solution. I work out a solution that serves my own as well as other's interestsas good as possible. I avoid a confrontationabout our differences. I avoid differences of opinion as much as possible. I try to make differences loom less severe. I try to avoid a confrontationwith the other.

Compromising

Forcing
9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20.

Problem solving

Avoiding

fromDUTCH andwere Note. Itemscould be answeredon a 5-point scale (1 = not at all, to 5 = very much). Items aretranslated presentedin a randomorder. The lean version (Study 1 and 2) does not include the compromisingscale; the expandedversion (Study 3) does include the compromisingscale.

Copyright? 2001 John Wiley & Sons, Ltd.

J. Organiz.Behav. 22, 645-668 (2001)