1992 Assessing Political Group Dynamics...

8/8/2019 1992 Assessing Political Group Dynamics...

1/23

INTERPERSONAL RELATIONS AND GROUPPROCESSESAssessing Political Group Dynamics: A Test of the Groupthink Model

Philip E. Tetlock, Rand all S. Peterson, Charles M cGuire, Shi-jie C hang, and Peter FeldInstitute of Personality a nd Social Research, U niversity of California, BerkeleyThis study used the Group D ynamics Q sort (GDQS) to explore the empirical unde rpinnings andtheoretical logic of the groupthink model. Examination of 10 decision-making episodes revealedconsiderable historica l supp ort for Janis's (1982) classification of groups into groupthink and vigi-lant decision-making episodes. LISREL analysis, however, revealed less support for Janis's (1982)causal m odel of groupthink. N either group cohesiveness nor situational stress emerged as indepen-dent predictors of symptoms of groupthink. Structural and proce dural faults of the organizationdid, however, emerge as a potent predictor. Little support was found for Janis's (1982) suggestionthat the ill-fated M ayaguez and Iran rescue decisions were the product of groupthin k. The resultsillustrate the sym biotic relationship between social psychology and history. Historical case studiesallow for testing theories of group dynam ics; social-psychological concepts and research m ethodscan inform interpretations of historical events.

Most po litical decisions in the world today are the product ofa collective decision-making process. Hitlers and Stalinsleaders who possessed such enormous personal authority thatthey could brutally im pose their ecce ntric views on entire societ-iesare relatively rare. It follows th at u ndersta nding how na-tional leaders make decisions requires more than understand-ing the belief systems and motives of individual decisionmakers. One also needs to un derstan d the m atrix of social rela-tionships within which key decision m akers work. The charac-ter of this collective decision-making process can profoundlyshape public policy. As a result of group interaction, people mayadopt more extreme or more moderate policies than theywould have as individuals; they may be come more or less self-critical and appreciative of the need to craft contingency plans,and they may become more or less aware of trade-offs and ofthe need to revise preconceptions in response to new evidence(Forsyth, 1990; Moscovici, 1985; Myers & L amm , 1976; Tet-lock, 1985). One can make a strong p rim a facie case that howthis group decision-making process unfolds plays a critical rolein determining the fates of governments.

Irving Janis (1972,1982,1989) was the most influential advo-cate of strong process-outcome links in group decision mak-ing. In his work on "g roupthink," Janis (1982) argued that pol-This study was supported by funds from the Mac Arthur Foundationand the Institute on Global Conflict and Cooperation. We are espe-cially grateful for the assistance and advice of Irving Janis, TerryBusch, George Breslauer, Greg Embree, David Fogelsong, Gail Lapi-dus, John Gaddis, Mary O'Sullivan, Renee Vuoncino, and RichardAbrams. We also appreciate the constructive criticisms of six anony-mous reviewers.Correspondence concerning this article should be addressed to Phi-lip E. Tetlock, Institute of Personality and Social Research, 2150 Kit-tredge Street, Oxford Court Building, Room 2C , University of C alifor-nia, B erkeley, California 94720.

icy failures often resulted from a decision-making process sodefective tha t t he resulting policies "deserved to b efiascoes"(p.9). In this article, we explore the validity of this groupthinkanalysis with a new research instrumentthe Group Dy-namics Q sort (GDQS)that allows us to quantify and com-pare expert assessments of group dynamics in a wide range ofhistorical settings.The groupthink model maintains that pressures for unifor-mity and loyalty can build up within group s to the point wherethey seriously interfere with both cognitive efficiency andmoral judgment. In case studies of several political decisions,Janis (1982) traced th e imp act of such social pressures on deci-sion making. In each case, he documented similar antecedentsof groupthink. Group members were invariably motivated tomaintain each other's respect and felt inhibited about criticiz-ing each other and expressing dissenting opinions. Groupmembers were insulated from qualified outsiders and lackedsystematic procedures for seeking and evaluating new evidence.During deliberations, the group leaders promoted their ownviews and discouraged open-minded analysis of options. Fi-nally, all decisions were made in stressful situations in whichpolicy makers doubted they could find a better option than theone the group favored.Under these conditions, several "symptoms" of groupthinkemerged. Decision makers began to believe that the group

could not faila belief that encouraged excessive optimismand risk taking. Group members discounted warnings con-cerning the preferred solution and refused to question eitherthe inhere nt righteousness of the group or the inherent imm or-ality of the enemy. Group members self-censored personaldoubts, fostering an illusion of unanim ity. On those rare occa-sions when someone did express a deviant opinion, the groupquickly silenced the critic and resolidified the consensus. Fi-nally, self-appointed mind guards shielded the grou p from ex-ternal sources of dissonant information.Journal of Personality and Social Psychology, 1992, Vol. 63, No. 3, 403-425Copyright 1992 by the American Psychological Association. Inc. 0022-3514/92/S3.00

403


2/23

404 TETLOCK, PETERSON, McGUIRE, CHANG, FELDIn this social atmosphere, decision-making procedures fellfar short of ideal "rational actor" standards. In all six cases ofgroupthink, Janis (1982) found numerous signs of defective de -cision m aking, including (a) truncated consideration of alterna-tives and objectives (often, th e g roup discussed only the optioninitially favored by group members), (b) a failure to examine th erisks of the initially preferred choice, (c) a failure to reappraise

initially rejected alternatives, (d) poor search for relevant infor-mation, (e) biased processing of information, and (f) a failure towork out contingency plans in the event known risks mater-ialized.Janis (1982) contrasted these groupth ink decisions with twoexamples of vigilant decision making: the Marshall Plan andthe Cuban Missile Crisis. In both cases, the decision-makinggroups gave high priority to critical appraisal and open discus-sion of options. The policies developed within these groupswere based on careful analysis of the likely consequences ofmany options, with frequent attempts at proposing new solu-tions that maximized the advantages and minimized the disad-vantages of options already ana lyzed.The groupthink analysis has provoked numerous critiques(e.g., Fischhoff & Beyth-M arom, 1978; Longley & Pruitt , 1980;McCauley, 1989; Steiner, 1982; & t'Hart, 1991; Whyte, 1989).Critics have raised four broad ob jections: (a) the inadequacies ofthe case study method for hypothesis testing (the risk of selec-tive attention to evidence and the tem ptation tofitmessy histor-ical facts into neat theoretical categories), (b) the suspiciouslyperfect correlation between soundness of process and goodnessof outcome (the risk of certainty of hindsight), (c) the all-or-nothing placement of decision-making episodes into thegroupthin k and vigilant categories (the risk of downplaying dif-ferences within a classification and of exaggerating differencesbetween classifications), and (d) conceptual misspecification ofthe m odel (challenging the causalflowfrom two of the anteced-

ent conditions group cohesiveness and provocative situationto concurrence seeking).Although controversial, Janis's (1982) analysis has receivedenormous professional attention and is widely cited in text-books. It is all the more surprising therefore tha t there has beenso little systematic research to test both the empirical supportfor, and internal logic of, the gro upthink mode l. Existing workfalls into two categories: laboratory experiments and contentanalyses of archival docum ents.The laboratory studies have tested facets of groupthinktheory by factorially manipulating key antecedent conditionsand then observing the effects on social interaction and deci-sion m aking (Callaway, M arriott, & Esser, 1985; Flowers, 1977;Leana, 1985; Moorhead & Montanari, 1986). Flowers (1977),for example, found th at an "open-leadership style" that encour-ages the free exchange of ideas leads groups to suggest moresolutions and to use m ore information than does a "closed-lea-dership style." This finding is compatible with Janis's (1982)hypotheses concerning leader directiveness. Contrary to Janis's(1982)findings, hough, group cohesiveness did no t affect deci-sion m aking. Th is latter finding triggered controversy. Flowersargued th at Janis's (1982) theory required revision; Janis (1982)contended that Flowers's operational definitions of low cohe-siveness (subjects who were strangers) and high cohesiveness

(subjects who were acquaintances) correspond poorly to the in-tense cohesiveness that emerges in cohesive policymakinggroups whose m embers have worked together closely for yearsand face a common fate.Taking a different tack, T etlock (1979) applied standardiz edcontent analysis procedures to the pu blic statements of key de-cision makers in groupthink and vigilant cases. He used twocoding schemesintegrative complexity coding (Tetlock &Suedfeld, 1988) and evaluative assertion analysis (Osgood, Sa-porta , & Nunnally, 1956)to assess the integrative com plexityof officials' statements and their evaluations of in-groups andout-groups. Consistent with Janis's (1982) theory, groupthinkdecision makers were more simplistic than vigilant decisionmakers in discussions of policy issues and m ade more positivereferences to in-group symbols (the United States and its Allies).Inconsistent with Janis's (1982) theory, groupthink decisionmakers did not make more negative references to out-groupsymbols (Communist states). Tetlock's (1979) study has its ownserious limitations. Co ntent analysis of public rhetoric is a ques-tionable m ethod of drawing inferences abou t the private deliber-ations of political groups. Tetlock m ay have simply tappe d intodifferent strategies of political impression mana gemen t.In short, no method is perfect. Each approach case studies,controlled experiments, and content analyses of archival re-cordshas distinctive strengths and w eaknesses. In this spirit,we introduce a fourth m ethod , the GDQS , an effort to com binethe methodological benefits of case studie s (descriptive realismand complexity) and of content analytic approaches (rigor andstandardization).

GDQSAn ideal m ethod of assessing leadership groups would com-bine richness with rigor (Snyder, 1985) by allowing investigators

to (a) assess a wide range of attribute s of group functioning; (b)describe group dynamics in complex, nuanced, and idiograph-ically sensitive ways; an d (c) make systematic, quantitative com-parisons across different analysts' assessments of the samegroup, across assessments of the same g roup at different tim es,and across assessments of different groups. Case studies satisfythe first 2 criteria but fail to pass the third. Different re-searchers emphasize different aspects of group functioning a nduse different terminologies to de scribe what they thin k is hap-pening. One researcher might emphasize the personality of thegroup leader through a series of vivid anecdotes, an other m ightstress factional infighting within the group, and a third mightfocus on the degree to which group mem bers have lost contactwith important economic trends. A reader might incorrectlyconclude that the experts profoundly disagree when they havecompletely complementary perspectives. They merely high-lighted different aspects of a complex, but sh ared, reality.If it is difficult to gauge agreement amon g case studies of thesame group; it is well-nigh impo ssible to gauge whether similarprocesses are at work across case studies of different groups.One must cope no t only with theoretical an d stylistic variationsamong researchers but also with highly idiosyncratic variationsamong leadership groups. Here, we confront t he classic limita-tion of case studies. Such studies may b e elegantly conce ptual-


3/23

TESTING GROUPTHINK 405ized, rigorously researched, and beautifully written, but theydon't add up (Verba, 1967). We lack a systematic framework foraccumulating insights across case studies.The GDQS solves this problem by (a) providing a comm ondescriptive language for capturing expert assessments of leader-ship group dynamics and (b) creating a standardized metric forinterjudge and intergroup comparisons.Developing a Comm on Descriptive Language

The G DQS consists of 100 pairs of bipolar statements thatdescribe a broad range of attributes of political leadershipgroups (see Method section for more detail). Assessors rate thedegree to which one or the other statement in each pair is moredescriptive of the group. Consider the following:The grou p leader is insulated from criticism.versusThe group leader is exposed to a wide range of views and argu-ments.

Assessors can usually retrieve from memory a number ofincidents that point to one or the other conclusion. The GDQ Srequires assessors, however, to reduce these complex historicalimpressions to a scalar judgment on a 9-point scale (where 1indicates that the upper statement is extremely characteristican d 9 indicates that the lower statement is extremely character-istic). We lose some information in this conversion process, butwe do gain a com mon data language. We can now measure thedegree to which two readers draw the sam e inferences from ahistorical case study or the degree to which readers of differentcase studies draw the same inferences. In addition, of specialinterest here, we can answer th e following questio n: Do readersof independently con ducted case studies see the sam e signs ofgroupthin k o r vigilance that readers of Janis's (1982) case stud-ies see?

We can, moreover, minimize the information loss. The 100-item GD QS covers diverse aspects of group functioning, m anydirectly relevant to the groupthink m odel. These items allow usto assess virtually all the cognitive, emotional, interpersonal,and organizational manifestations of groupthink that Janis(1982) posited. The items fall into seven conceptually interre-lated and empirically intercorrelated clusters that we call pro-cess indicator scales (see Method section). Our theo retical ex-pectations were that readers of historical case studies wouldassign grou pthink episodes substantially higher scores than vig-ilant episodes on three process indicator scales: (a) LeaderStrength (high scores indicate assertive leaders who push the irpreferred solutions and clearly com munica te what they w ant),(b) Rigidity (high scores indicate that the group is insulatedfrom qualified outsiders and e ncourages self-righteous and dog-matic styles of thinking ), and (c) Conformity (high scores indi-cate that group members actively discourage dissent and cen-sure deviants). In additio n, we expected Q-sort assessors to as-sign group think cases lower scores than vigilant cases on fourother scales: (a) Factionalism (low scores indicate a cohesivegroup w hose members share well-defined goals and who rarelydisagree), (b) Democratic Accountability (a low score indic ates

a group that has contempt for the n orms of democratic gover-nance), (c) Pessimism (a low score indic ates collective self-confi-dence), and (4) Task Orientation (a low score indicates thegroup devotes more energy to ensuring tha t mem bers feel goodabout them selves than t o working out viable solutions to policyproblems).Developing a Standardized Metric

The GDQS requires assessors to assign a certain number ofcards to each of the nine categories to create a quasi-normaldistribution of cards, with the most cards in Category 5 andgradually fewer cards as one moves toward the extremes of oneand nine. Two advantages follow from this ipsative scaling ofjudgments. First, the Q sort requires observers to judge therelative descriptiveness of each item a nd to make difficult inter-item comparisons (e.g., is Item 6 really more descriptive thanItem 47?) that can be avoided by using simpler rating scalemethods. Although such comparisons are time consuming, theQ sort yields higher quality data tha n do less demanding meth-ods (Block, 1978). Second, and a related point, by creating acomm on measurement scale, the G DQS eliminates irrelevantvariation in judgmental style (e.g., the tendency of some ob-servers to make either middle-of-the-road or extreme judg-ments). Wh en two observers report different Q sorts, we can bereasonably confident tha t they are doing so not merely becausethey use the rating scale differently but because they see thegroup differently (Block, 1978).To summ arize, our approach to assessing historical supportfor groupthink theory involves three steps: (a) asking theoreti-cally neutral assessors to describe groupthink and vigilantgroups using the Q sort; (b)asking assessors to generate these Qsorts from a variety of historical sources for each of the eightcases that Janis (1982) initially examined; and (c) comparing

these Q sorts with one another, with Q sorts derived from Janis's(1982) case studies, and with theory-derived ideal ty pe Q sorts(cf. Bern & Funder, 1978, on tem plate m atching).Our study also had ad ditional objectives. The Q-sort methodallows us to study group s far beyond the highly selective sam-pling of cases in the Janis (1982) study, which purposely focusedon extreme cases of both groupthink and vigilance. Assumingthis classification holds up against Q sorts derived from otherhistorical sources, it is reasonable to ask whether new cases fitalready docum ented p atterns of either groupthink o r vigilance.For this study, we examined two decision-making episodes thatJanis (1982) did not formally classify (although he did proposethem as candidates for groupthink): the Ford administration'sMayaguez rescue decision and the Carte r administration's Iran

rescue decision. There are at least three possible outcomes.First, if the Mayaguez and Iran rescue cases match the group-think pattern, we will have further evidence of the pervasive-ness of this "syndrom e" in political decision making. Second , ifthe new cases fall into th e vigilant classification, we w ill haveidentified important exceptions to the generalization that thequality of decision-making procedures and goodness of out-come covary closely in the re al world (both th e Mayaguez andIran rescue decisions are widely classified as failures). Third, ifthe new cases fall somewhere in the middle of the groupthink-


4/23

406 TETLOCK, PETERSON, McGUIRE, CHANG, FELDvigilance continuum, we will have a useful reminder of theneed to develop more nuanced and differentiated classifica-tions of decision making.

Finally, the current study allows us to test the causal assump-tions underlying groupthink theory. It is logically possible, forexample, that historical cases resemble each other or differ inapproximately the ways specified by Janis (1982) but that thepatterns of covariation among Q-sort items diverge markedlyfrom the causal model that Janis (1982) specified for group-think. This model makes strong claims: It identifies multiplenecessary causesCohesiveness, Structural and ProceduralFaults of the Organization, and Provocative Situational Contextfor the social and cognitive symptoms of groupthink toemerge. The model also makes strong mediational claims: Thehypothesized antecedents directly influence only the tendencytoward concurrence seeking, which, in turn, generates socialsymptoms of groupthink, which, in turn, lead to defective deci-sion making. The specified antecedents of groupthink may,however, be neither necessary nor sufficient conditions for con-currence seeking. The specified antecedents may also producesocial symptoms of groupthink or defective decision making,but not through the hypothesized mediator of concurrenceseeking. Simpler theoretical modelssuch as those of Longleyand Pruitt (1980) and Steiner (1982)may better capture thecovariations among items. These models raise a host of ques-tions that we can explore by drawing on the LISREL (Joreskog& Sfirbom, 1989) method of causal modeling: Is group cohesive-ness necessary to produce concurrence seeking? Are proce-dural faultssuch as leader directivenessthe driving causebehind whetherflexible,self-critical versus rigid, self-righteousthinking emerges in decision-making groups? Are situationalstressors necessary for groupthink or are they quite irrelevant?

MethodInstrument Development

In generating items for the Q so rt, we drew on diverse sources, in-cluding experimental social psychology, organizational behavior, andboth contemporary and historical studie s of leadership practices in abroad ba nd of political cultures. On t he basis of this review, we identi-fied a number of critical dimensions of group functioning that neededto be represented in the GD QS item pool. These dimensions included(a) properties of group leadership (authority and sta ture w ithin group,style of political op erati on, activity level, cognitive style, and interpe r-sonal style), (b) the group's decision-making culture (tolerance for dis-sent and self-critical policy analysis, attitudes toward risk, optim ism -pessimism, importance of achieving full consensus, collective decisionrules, and the amount of secrecy surrounding the group and its func-tioning), (c) the coalitional or factional struc ture of the group (the na-ture and intensity of intragroup rivalries), (d) the dominant world viewof the grou p (widely shared assu mptio ns about the group and its role inthe worldassumptions that serve as premises for action), (e) groupcomposition (homogeneity of group, key attributes ofgroup members,such as self-esteem, defensiveness, and need for approval), (f) patron-age networks (pervasiveness of special privileges and corruption andexchange relationships between leaders a nd group members), (g) insti-tutional and constitutional constraints (accountability of th e group tothe law and legislative bodies), and (h) types and severity of challenges

confronting the group (internal challenges, external threats, and politi-cal-economic pressures).To ensure that we had comprehensively sampled the universe ofitems descriptive of political group dynam ics, we pretested an initialset of 300 items on a group of 20 political scientists, historians, andleadership analysts. These experts applied the 300-item version to suchdiverse groups as the cabinets of British Prim e M inisters (BenjaminDisraeli/1874, James Callaghan/1979, and Margaret Thatcher/1989),Soviet politburos (Joseph Stalin/1953, Nikita Khrushchev-GeorgyMalenkov/1954, Leonid Brezhnev/1965, and Konstantin Cherne nko/1984), the European C ommission (1989), and th e inner decision-mak-ing circles of the leaders of North Korea (Kim II Sung/1989), Equato-rial Guinea (Colonel Nguema in 1979), South Africa (F. W deKlerk/1989) and Romania (Nicolae Ceausescu/1988). On the basis of itemintercorrelations and experts' comm ents, we dramatically reduced thelength of the Q sort to 100 items and retested the revised instrume nt ona more extensive sampling of groups (approximately 30). Experts werevirtually unanimous that the instrument captured the key patterns ofleadership group dynamics in the periods with which they were famil-iar. Appendix A presents a complete listing of Q-sort items.

We derived the seven process indicator scales by analyzing the inter-correlations among all items across all groups that were not part of thegroupthink study reported here. Cronbach alphas ranged from .57to .91.Process Indicator Scales

The p rocess indicator scales provide an efficient mean s of summar-izing a Q sort. These scales included (a) Leader Strength (Items 24,32 ,39,44,60, and 83), (b) Factionalism (Items 7,9 reversed, 11,4 1,51 ,53reversed, 72 reversed, 78, an d 86), (c) Rigidity Index (Items 2 reversed,12,19, 29, 37, 58 reversed, 74 reversed, 81 , 82 reversed, and 88), (d)Democratic Accountability (Items 4 reversed, 5 reversed, 6, 10 re-versed, 16,22,28,33 reversed, 52,65,93 reversed, and 95 reversed), (e)Pessimism (Items 17,55,64 ,60,93, and 96), (0 Conformity (Items 1,14reversed, 40, 49 reversed, 53, and 66 reversed), and (g) Task O rienta-tion (Items 35, 43 reversed, 66, 74, 88 reversed, and 90 reversed).Theoretical Ideal Types

The ideal type s were created by Philip E. Tetlock and two researchassociates who drew on the historical and theoretical literature to iden-tify 12 extreme forms that leadership group dynamics can take.1 Ap-pendix B lists extreme item rankings for the 12 ideal types. Thesetheoretical Q-sort profiles could then be matched against actualgroups to gauge the resemblance to particular templates (cf. Bern &Funder, 1978). The first step was to draw u p a brief description of thedefining features of the ideal typ e. For instance, the ideal typ e ofklep-tocracy was defined by an "exclusive concern am ong m embers of th eleadership group with their personal enrichment and the enrichmentof key supporters, an absence of institutional and legal constraints onthe group, and a lack of interest in details of governance except insofaras they impinge on prospects for self- or group aggrandizement." Thehistorical examples of the Ferdinand Marcos regime in the Philip-pines and the Francois Duva lierregime n Haiti were provided (basedon nominations from a panel offivespecialists in comparative poli-tics). The second step was todescribe each ideal type using all 100hems of the GD QS and to create a composite of the 3 independent Qsorts (average interrater agreement, r = .74).

1 We acknowledge the assistance of the late Irving Janis in construct-ing the Q-sort template for the ideal type o f groupthink.


5/23

TESTING GROUPTHINK 407The 12 ideal types include absolutism (authority and power are t o-tally centralized in one person w ho claims to embody the w ill of thepeople and all political wisdomStalin, Mao Ze-Dong, and Kim IISung), kleptocracy (denned earlier), competitive oligarchy (a blend ofauthoritarian centralization of power in a small group and tendenciestoward factionalism and power struggles among key actors), cohesiveoligarchy (authoritarian centralism plus esprit de corps and mutualcooperation among group members), stable democratic polity (thegroup works comfortably within well-established c onstraints of a dem-ocratic system), groupthink (defined earlier), incipient anarchy (col-lapse of internal and external authority and legitimacy), technocracy (agroup of experts who work hard a nd seek technical-scientific solutionsto problems of governance, even if that offends traditional sensitivi-ties), multiple advocacy (based on work of Alexander George, 1980,the leader encourages free policy debate but main tains clear controlover ground rules and thefinaldecision), group in systemic crisis (thisideal type focuses mostly on the political-eco nomic environm ent con-fronting the grouplots of painful trade-offs), demoralized group(group members have disengaged from their jobs, and apathy and dis-illusion reign), and collective madness (here group dynamics and psy-chopathology merge: Unde r intense stress and the control of a b izarrepersonality, the group becomes de tached from realitya classic histor-ical example is the fantasy w orld into which Hitler a nd associates re-

treated in the Berlin bunker in April, 1945).Assessing Key Components of the Causal ModelofGroupthink

Janis (1982) identified three categories of antecedent conditions andthree categories of consequences of groupthink. We used the followingitems to operationalize the antecedent cond itions: (a) Cohesiveness ofDecision Makers included Items 11, 23, 34, 55, and 99; (b) StructuralFaults of the Organization included Items 2, 6, 8 reversed, 9 reversed,10 reversed, 12 reversed, 14,28,3 0,36 reversed, 39 reversed, 47,52,53reversed, 81 reversed, 82,83 reversed, and 92 ; and (c) Provocative Situa-tional Context included Items 26,2 7, 31 reversed, 38 reversed, 56, 59reversed, 71 reversed, and 76. We used the following items to o pera-tionalize m anifestations and consequences of groupthink: (a) Concur-rence-Seeking Tendency included Items 1 reversed, 4 reversed, 7, 24reversed, 40 reversed, 60 reversed, and 66; (b) Social Symptoms ofGroupthink included Items 3 reversed, 17,18 reversed, 22 reversed, 29reversed, 37 reversed, 64,68 reversed, 70 reversed, 88 reversed, and 94;and (c) Symptoms of Defective Decision Making included Items 19reversed, 58, 74, 77 reversed, and 98.Janis (1982) Baseline Q Sorts

The first step in testing the groupthink model was to translate theJanis (1982) case studies of the six groupthink and two vigilant deci-sion-making episodes into the standardized data language of theGDQS . The groupthink cases included the appeasem ent decisions ofthe Neville Cham berlain cab inet (1938); the lack of preparedness ofAdmiral Husband Kimmel's command for the Japanese attack onPearl Harbor in December, 1941; the decision to pursue the defeatedNorth Korean army beyond th e 38th parallel (October, 1950); the deci-sion of the Kennedy administration to launch the Bay of Pigs invasion(April, 1961); the Johnson administration's decisions to escalate in-volvement in the Vietnam War (spring, 1965); and the attem pted cover-up by the Richard Nixon Watergate group (most of 1973). The vigilantcases included the dev elopment of the Marsha ll Plan (spring, 1947) andthe Kennedy administration's handling of the Cuban Missile Crisis(September-October, 1962). At least 2 Q sorters read each case studyand w ere instructed to portray the g roup in question as Janis (1982)

had in the case study (not as the Q sorter h im- or herself believed th egroup to have been).

Q-Sort Assessments From Other SourcesTo test the groupthink model, we drew on the voluminous historicalliterature on each of the eight m ajor decision-making cases that Janis(1982,1989) examined. For each of these cases, we identified the majorbooks and articles published by historians and political scientists onthe decision-making dynamics of the leadership groups. We identifiedat least three major historical sources for each decision-making case(drawing on the recomm endations of experts and on citation patte rnsin th e literature). Two independent observers (one of whom was alwaysunaware of the hypotheses being tested) carefully read each sourceand, based on the information contained within that source, com-pleted a Q sort that captured as closely as possible the portrayal of thegroup in the source. A total of 9 observers (graduate students andpostdoctoral fellows) served as Q sorters in this study.We also obtained Q-sort assessments of two decision-making epi-sodes that Janis (1982) speculated might have involved groupthink but

did no t thoroughly investigate: the Ford administration's decision t orescue the crew of the Mayaguez from captivity in Kampuc hea (May,1975) and the C arter adm inistration's decision to rescue the U.S. hos-tages from t he embassy in Tehran (April, 1980). Again, a minimum of 2Q-sort assessors read a minimum of three historical case studies ofeach decision.Table 1 presents the texts for each historical case, th e key members ofeach group, and the tim e frame for the Q-sort assessment. We selectedonly those texts that satisfied two or more of the following criteria: (a)All sources had to contain enough descriptive detail to perm it hypoth-esis-blind GDQ S assessors to perform a Q sort of the group (as a result,books and articles that focused on other levels of analysise.g., strate -gic or geopoliticalwere excluded), (b) the source was recom mendedby a panel of four political scientists and historians known for theirwork on American foreign policy, and (c) the source was listed in thewidely used reference source for diplomatic historians the G uide toAmerican Foreign Relations.

Q-Sorting ProcedureWe obtained two or three Q sorts that were based on each historicalsource listed in Table 1. Q sorters were always reminded that th eir taskwas not to portray the group the way they thought it was but rather t oportray the group as the auth or of the assigned text believed it to be . Qsorters also received the standard procedural instructions:

Th e Q-sorting procedure is simple, but somewhat tim e consum-ing. With the group to be assessed in m ind, look through the deckof 100 cards. You will note tha t each card has an upper statementand a lower statement th at are opposites. First, sort the cards intothree stacks in a row. Place in th e up per stack all those ca rds forwhich the upper statement is characteristic of the group. Place inthe lower stack all those cards for which the lower statement ischaracteristic of the group. Place in the middle the remainingcards. No attention need be paid to the num ber of cards falling ineach grouping at this time.When the three stacks have been established, they must be fur-ther fractionated into a row of categories, placing the m ost charac-teristic statements at the two ends of the row [as shown in Table 2 ].\bu may feel frustrated by the constraints of the sorting proce-


6/23

408 TETLOCK, PETERSON, McGUIRE, CHANG, FELDTable 1Historical Cases and Texts Used

Group Major actors TextsChamberlain cabinet(June 2, 1937-May 10, 1940)

Truman cabinet(August-October, 1950)

Kennedy cabinet:Bay of Pigs(January-April,1961)

Johnson cabinet(December,1964-July,1965)

Pearl Harbor:Kimmel(February 1,1941-December 7,1941)

Neville Cham berlain, LordHalifax, Sir JohnSimon, Sir SamuelHoare, Sir AlexanderCadogan, Sir HoraceWilson, Sir RobertVansitart

President Truman, DeanAcheson, LouisJohnson, GeorgeMarshall, GeneralDouglas MacArthur,General Collins,General Bradley,General Vendenberg,Admiral ShermanPresident Kennedy, AllenDulles, Richa rd Bissel,Ted Sorenson, ArthurSchlesinger, DeanRusk, RichardGoodwin, McGeorgeBundy, RichardHelms, Jack Engler,Robert McNamara,Walt Rostow, GeneralGray, GeneralLemnitzer, AdmiralBurkePresident Jo hnson, VicePresident H.Humphrey, DeanRusk, RobertMcNamara, John

McCone, MaxwellTaylor, General E.Wheeler, McGeorgeBundy, WilliamBundy, JohnMcNaughton, GeorgeBallAdmiral H. KimmelCaptain W. SmithRear A dmiral P . BellingerVice Admiral W. PyeCaptain C. McMorrisVice Admiral W. Halsey

Nixon group(January, 1972-April, 1973)

Lieutenant Commander E.LaytonRear A dmiral C. BlochPresident Nixon, H. R.Haldeman, JohnEhrlichman, JohnDean, John Mitchell,Charles Colson, JebMagruder, DwightChapin, RichardKleindienst

Colvin, I. G. (1971). Th e Chamberlain cabinet: How the meeting in 10 DowningStreet, 1937-1939, led to the Second W orld War. London: Gollancz.Fuchser, L. W. (1982). Neville Chamberlain an d appeasement: A study in thepolitics of history. New York: Norton.Hyde, H. M. (1976). Neville Chamberlain. London: Weidenfeld & Nicolson.Middlemas, K. (1972). Diplomacy of illusion. London: Weidenfeld & N icholson.

Lichterman, M . (1963). To the Yalu and back. In H. Stein (Ed.), Americancivil-military decisions: A book of case studies (pp. 471-491). Birmingham:University of Alabama Press.Donovan, R. J. (1984). Nemesis: Truman and Johnson in the coils of war inAsia. New York: St. Martin's M arek.Kaufman, B. (1986). Th e Korean war: Challenges in crisis, credibility, andcommand. Philadelphia: Temple University Press.

Wyden, P . (1979). Bay of Pigs. The untold story. New York: Simon & Schuster.Vandenbrouke, L . (1984). Anatomy of a failure: The decision to land at the Bayof Pigs. Political Science Quarterly, 99, 471-491.Schlesinger, A., Jr. (1965). A thousand days: John F. Kennedy in the WhiteHouse. New York: Fawcett Premier.

Donovan, R. J. (1984). Nemesis: Truman and Johnson in the coils of war inAsia. New York: St. M artin's-Marek.Karnow, S. (1983). Vietnam: A history. New York: Viking Press.Graff, H. F. (1970). The Tuesday cabinet: Deliberation an d decision on peacean d war under Lyndon B. Johnson. Englewood Cliffs, N J: Pren tice-Hall.

Prange, G. W . (1986). Pearl Harbor: Th e verdict of history. New York:McGraw-Hill.Utley, J. (1985). Going to war with Japan: 1937-1941. Knoxville, TN :University of Tennessee Press.Morison, S. E. (1950). History of U.S. naval operations in World War II. Boston:Little, Brown.Wohlstetter, R. (1962). Pearl Harbor: Warning and decision. Stanford, CA:Stanford University Press.Kutler, S. 1. (1990). Th e wars of Watergate: The last crisis of Nixon. New York:Knopf-Random HouseRaven, B. H. (1974). The Nixon group. Journal of Social Issues, 30, 297-320.Ehrlichman, J. (1982). Witness to power: the Nixon years. New York: Simon &Schuster.


7/23

TESTING GROUPTHINK 409Table 1 (continued)

Group Major actors TextsMarshall Plan (May,1947-December,1947)

Kennedy: CubanMissile Crisis(September-October, 1962)

Ford: Mayaguezrescue (May,1975)

Carter: Iran hostagerescue attemp t(November,1979-April,1980)

George Marshall, DeanAcheson, RobertLovett, WilliamClayton, CharlesKindleberger, GeorgeKennan, (Kennanstaff: Charles Bohlen,Carlton Savage, JohnDavies, JosephJohnson, BurtonBerry, Jam es Angell,George McGhee, andWeir Adams)

President Kennedy,Robert Kennedy,Dean Rusk, RobertMcNamara, JohnMcCone, DouglasDillon, McGeorgeBundy, Ted Sorenson,George B all, U. AlexisJohnson, General M.Taylor, EdwardMartin, LlewellynThompson, RoswellGilpatrick

President Ford, HenryKissinger, JamesSchlesinger, B rentScowcroft

President Carter, ZbigniewBrzezinski, StansfieldTurner, GeneralVaught, GeneralJones, Cyrus Vance,Hamilton Jordan

Gimbel, J. (1976). Th e origins of the Marshall Plan. Stanford, CA: StanfordUniversity Press.Pogue, F. C. (1987). George C. Marshall, statesman, 1945-1959. New York:Viking Press.Mee, C. L. (1984). The Marshall Plan: The launching of Pax Americana. NewYork: Simon & Schuster.Hogan, M . J. (1987). Th e Marshall Plan: American, Britain, and thereconstruction of Western Europe. Cambridge, England: Cam bridgeUniversity Press.

Blight, J. G., & Welch, D. A. (1989). On the brink: Americans and Sovietsreexamine th e Cuban Missile Crisis. New York: Hill & Wang.Kennedy, R . (1969). Thirteen days: A memoir of the Cuban Missile Crisis. NewYork: Norton.Garthoff, R. (1989). Reflections on the Cuban M issile Crisis. Washington DC:Brookings Institution.

Lamb, C . (1981). Belief systems and decision making in the Mayaguez crisis.Gainsville, FL: University of Florida Press.Ford, G. (1979). Time to heal. New York: Harper & Row.Head, R . G., Short, F. W., & M cFarlane, R. C. (1978). Crisis resolution:Presidential decision-making in the Mayaguez and Korean confrontations.Boulder, CO: Westview Press.Brzezinski, Z. (1983). Power and principle. New York: Farrar, Straus & Giroux.Jordan, H. (1982). Crisis: The last year of the Carter presidency. New York:Putnam.Vance, C. (1983). Hard choices. New York: Simon & Schuster.Ryan, P. B. (1985). Th e Iranian rescue mission: Why it failed. Annapolis, MD:Naval Institute Press.

dure. In justification, it should be noted tha t specifying the num-ber of cards to be assigned to each category has proven to be amore valuable procedure than the freer situation in which thejudge can assign any number of cards to a category. Past researchindicates that we underestimate the degree of interjudge agree-ment when there are n o constraints on sorting. The reason is sim-ple. When we compare tw o free-form Q sorts, there are threecauses of disagreem ent a t work: real differences in poin t of view,random error variance (mood, carelessness, . . . ), and differ-ences in how the judges use the rating scale (we know that somepeople make extreme judgm ents, whereas others are fence sitters).When we compare two forced-distribution Q sorts, we eliminatethis thi rd source of interjudge disagreement (really pseud o-dis-agreement) by standard izing how everyone uses the rating scale.This forced-distribution Q sort also has another related advan-tage. Because it limits the number of items per scale value cate-gory, the forced-distribution Q sort puts pressure on judges tomake frequent comparisons of herelativedescriptive appropriate-ness of items. It is possible to highlight only so many items in the"extremely characteristic" categories. As aresult, here is a need toset descriptive p riorities in assessing a group. One m ust ask one-self the following: Given that I can highlight only a handful ofstatements as extremely characteristic, which ones are particularly

worthy of being singled out? Making comparisons of t his sort isnot easy, but it does increase both the interjudge reliability andpredictive value of Q sorts.

ResultsReliability of the Q-Sort Assessments

We computed three measures of Q-sort reliability: (a) in-terrater agreement (do readers of the same historical texts drawsimilar conclusions?), (b) intertext agreement (do different his-torical accounts lead readers to similar conclusions?), and (c)internal consistency of process indicator scales (do concep-tually related items intercorrelate?). With respect to interrateragreement, the average correlation was .83, with a range from.42 to .94 (based on 45 interrater com parisons). These reliabil-ities compared favorably with widely used psychological testsand justified proceeding to the next level of analysis: collapsing


8/23

410 TETLOCK, PETERSON, McGUIRE, CHANG, FELDacross raters to create composite Q sorts for each tex t. Intertextagreement was lower but still substantial. We computed 33correlation coefficients between Q-sort composites derivedfrom different textual perspectives on the same group. Theaverage correlation was .52, with a range of. 17 to .88.2 This levelof intertext agreement justified collapsing composite Q sortsacross texts to create supercomposites that could b e comp areddirectly with Q sorts derived from Janis's (1982) model and casestudies.

The average Cronb ach alp ha coefficient for the seven processindicator scales was .72. Individual coefficients were .66 forLeader Strength, .70 for Factionalism, .88 for Rigidity, .71 forDemo cratic Accountability, .57 for Pessimism, .91 for Confor-mity, and .61 for Task Orien tation.Th e average alpha coefficient for the m easures used inLISREL m odeling was .75. Individual alphas were .79 for C ohe-siveness of Decision Makers, .81 for Structural Faults of theOrganization, .61 for Provocative Situational Context, .69 forConcurrence-Seeking Tendency, .74 for Social Symptoms ofGroupthink, and .90 for Symptoms of Defective Decision

Making.3

Correlations Among Process Indicator ScalesTable 3 presents the correlations among process indicatorscales. The average correlation am ong scales was .35. In a fewinstances, the correlations were strikingly high (in particularamong Rigidity, Conformity, and Task Orientation). Thesecorrelations largely reflect the universe from which we sampledgroups. Groupthink theory led us to expect that groups thatscored high or low on one scale would do the same on the

Table 2Q-Sorting Procedure an d Distribution Constraints

Table 3Process Indicator Scale Intercorrelations (N = 84 )

Category

Statement1234

Category5

Statement67g9

LabelUpper

Extremely characteristicHighly characteristicQuite characteristicSlightly charac teristicMiddle

Neither upper n or lowerstatement is characteristicLower

Slightly characteristicQuite characteristicHighly characteristicExtremely characteristic

No . ofcards

581216

18

161285

Scale1. LST2. FAC3. RIG4. DEM5. PES6. CON7. TO

1_.3 0.1 0.22.21.4 2.0 9

2

.4 5

.1 5.0 3.52.3 8

3

.3 6.0 5.8 2.9 0

4

.6 6.3 6.4 4

5

.1 0.0 5

6

.78

7

Note. LST = leader strength; FAC = factionalism; RIG = rigidity;DEM = dem ocratic accountability; PES = pessimism; CON = confor-mity; TO = task orientation.

others. Had a wider variety of groups b een assessed, we wouldhave expected lower correlations.Assessing the Historical Support for the GroupthinkModelThe groupthink model posited several indicators of concur-rence seeking and defective decision-making process. Table 4reports the process indicator scales most relevant to testingthese predictions (scores derived by averaging all Q sorts foreach group). The first 6 groups are hypothesized groupthinkcases and the last two are hypothesized cases of vigilance.Figure 1 presents the seven process indicator scores (col-

2 This reliability measure con tains two sources of errorinterraterand intertext disagreement. Given the high level of interrater agree-ment reported earlier (.83), we attribute the reduced reliability prima r-ily to disagreement among the case studies.3 Additional evidence on the construct validity of both the idealtypes and process indicator scales is available in our database. Forinstance, experts on the Soviet political system have provided Q sortsof Politburo group dynam ics at the following junctu res in Soviet his-tory: the late Stalin Politburo (January, 1953), the Khrushchev-Malen-kov politburo (of 1954), the early Brezhnev (of 1965) Politburo, theChernenk o Politburo (of July, 1984), and a nu mber of Q sorts of theGorbachev Politburo (between 1985 and 1990). Examination of theideal type correlates of these Q-sort descriptions of actual leadershipgroup dynamics reveals a pa ttern c onsistent with what Soviet expertscall the "changing bases of power" in the Soviet political system. Wemove from a reliance primarily on coercion, intimidation, and terrorin the late Stalin period (e.g., a strikingly high .75 correlation withabsolutism ideal type and strikingly high scores on the Coercion pro-cess indicator scale), to collective leadership by a small but deeply

divided oligarchy (a dramatic reduction in the correlation with theabsolutism and a d ramat ic increase in the correlation with the com peti-tive oligarchy ideal type, .60), to increasing corruption and patronageduring the Brezhnev period up to Chernenk o (the highest correlatingideal type now emerges as kleptocracy, r = .57), and finally to theemergence of group norms of democratic and constitutional account-ability in the Gorbachev era (with substantial correlations emergingwith the ideal types of multiple advocacy and democratic polity, ap-proximately .40). These correlations are perhaps not surprising, butthey provido independent evidence of the sensitivity of the instrumentto well-documented historical trends.


9/23

TESTING GROUPTHINK 411Table 4Mean Process Indicator Scores fo r Group Composite Q Sorts

GroupChamberlainTrumanKennedy (Bay of Pigs)LBJKimmelNixonMarshallKennedy (missile crisis)

LST6.95.05.96.66.37.25.05.7

FAC4.34.64.54.03.84.55.85.2

RIG6.66.36.65.06.36.13.13.2

DEM5.95.76.35.56.04.55.96.3

PES3.73.43.94.63.05.23.84.5

CO N7.06.27.05.46.67.13.03.0

TO5.24.24.55.35.34.56.47.3

n86668697

Note. Scores are collapsed across texts, excluding Janis (1982). The higher the score, the greater theattribute. LST = leader strength; FAC = factionalism; RIG = rigidity; DEM = dem ocratic accountability;PES = pessimism; CON = conformity; TO = task orientation; LBJ = Lyndon B. Johnson.

lapsed over raters, texts, and h istorical cases) for the groupthinkand vigilant decision-making cases. A one-way multivariateanalysis of variance (MANOVA) yielded significant differencesbetween the groupthink and vigilant groups, Wilks's .F(10,43) = 25.80, p < .001. Discrim inant analysis clarified the rela-

tive impo rtance of the process indicator scales in differentiat-ing the two categories of groups (standard canonical coeffi-cients in parentheses): Rigidity (.93), Conformity (.60), LeaderStrength (.56), Factionalism (-.54), Dem ocratic A ccountability(.33), Pessimism (.26), and Task Orientation (-. 15). Overall, the

Proce

IndieatorScores

2 -

LST FAC RIG DE M PSM CO N T -OG roup th i nk G roups V i g i l an t G roups

Figure 1. Process indicator scores for the combined Q sorts of all groupthink and vigilant cases derivedfrom historical studies. (LST = leader strength; FAC = factionalism; RIG = rigidity; DEM = democraticaccountability; PSM = pessimism; CON = conformity; T-O = task orientation.)


10/23

412 TETLOCK, PETERSON, McGUIRE, CHANG, FELDdiscrim inant function was significant at the .0001 level, x2 (10,N = 74) = 136.9, and allowed us to predict the groupthink orvigilant origins of 99% of the classified Q sorts (against a chanceaccuracy of 75%).We followed u p these multivariate tests with analysis of vari-ance (ANOVA) planned co mpa risons th at revealed su bstantialdifferences on five of the seven scales. Groupthink episodesshowed (a) greater Leader Strength, F(l, 52) = 11.07, p < .002,w2 = .16, M S = .87; (b) less Factionalism, F (l, 52) = 35.85 , p

1992 Assessing Political Group Dynamics...

Documents

Transcript of 1992 Assessing Political Group Dynamics...