Human Relations, Volume 39, Number 5, 1986, pp. 399-410 An Empirical Investigation of the Groupthink Phenomenon! Gregory Moorhead and John R. Montanari Arizona State University This paper reports results of an empirical investigation of the groupthink theoretical framework presented in Janis’ second edition of Groupthink. Fac- tor analysis was used to develop scales to measure the aspects of groupthink proposed by Janis (1982). Results supported some, but not all, of the symp- toms and defects postulated by Janis. Next, a path analytic procedure was utilized to test the implied causal ordering of the Groupthink model (Janis, 1982, p. 244). This study provides limited support for the causal sequence specified in the model. However, the relationship between groupthink-induced decision defects and outcomes were not as strong as Janis suggests. These results prompt the authors to agree with Courtright (1978) that many interven- ing and/or moderating factors not included in the Janis framework influence decision outcomes. INTRODUCTION The study of group decision-making and problem-solving processes has been a major focus of the organizational behavior literature for several decades (Vroom, Grant, & Cotton, 1969; Cartwright, 1971). In 1972, Janis contributed to this area of study when he coined the term “groupthink” to refer to “a mode of thinking that people engage in when they are deeply in- volved in a cohesive in-group, when the members’ strivings for unanimity over- ride their motivation to realistically appraise alternative courses of action” (p. 8). Support for the presence of this phenomenon was based on his ’Portions of this paper were presented at the 1Sth Annual Meeting of the American Institute for Decision Sciences, San Antonio, Texas, November 1983. *Requests for reprints should be sent to Gregory Moorhead, Department of Management, Arizona State University, Tempe, Arizona 85287. 399 267/86/0500-0399S05.00/1 © 1986 Tavistock Institute of Human Relations from the SAGE Social Science Collections. All Rights Reserved,400 Moorhead and Montanari historical analysis of the decision-making activities of governmental policy- making groups that resulted in either major “fiascoes” or outstanding suc- cesses. Although his conclusions represent a major departure from well- established concepts of the performance of highly cohesive groups, they have seldom been totally supported by empirical analysis. Perhaps one reason for this dearth of research on the groupthink phenomenon is the failure of Janis to develop a concise theoretical framework in his original book. In the 1982 version of Groupthink, Janis presents an illustrated model and discussion of a theoretical framework that permits a more precise interpretation of the causal sequence alluded tu in his 1972 work. The primary objective of this research was to empirically investigate the con- cepts that are proposed in the groupthink model and their implied causal sequence. However, in order to accomplish this goal, it was necessary to operationally define and develop valid measurement scales for the group- think variables. The primary thesis of the framework proposed in Janis’ (1982) model is that the presence of specific antecedent conditions increases the likelihood that the group will develop attitudes and behaviors symptomatic of group- think. Furthermore, these symptoms promote observable defects in the group’s decision-making processes which result in poor quality decisions. That is, the decision-making procedures in the fiascoes were faulty, regardless of the outcomes. Likewise, the decision-making procedures in the successes were appropriate, regardless of the outcomes. This emphasized the distinction bet- ween decision-making procedures and decision quality. However, Janis con- tends that the probabilities for successful outcomes, given a faulty decision-making process, are very low. In his original work, Janis (1972) identified six major defects in the decision-making process which result from groupthink: (1) discussions were limited to only a few alternatives, (2) the originally preferred solution was not reevaluated, (3) alternatives which were initially discarded were never reevaluated, (4) the advice of experts was not sought, (5) where advice was presented, selective bias on the part of members was evident, and (6) members failed to consider how groups external to the focal group might react and failed to develop contingency plans. The more recent edition includes a seventh defect, incomplete survey of objectives. These defects result in poor quality decisions and are related to the presence of strong concurrence-seeking antecedent conditions associated with the groupthink phenomenon. The primary antecedent condition necessary for groupthink is a highly cohesive group. Secondary conditions that relate to structural faults of the organization are: (1) insulation of the group, (2) lack of a tradition of im- partial leadership, (3) lack of norms requiring methodical procedures, and (4) homogeneity of members’ social background and ideology. TertiaryThe Groupthink Phenomenon 401 antecedent conditions relate to the decision-making context and include high stress from external threats with low hope of a better solution than the leader's and low self-esteem temporarily induced by group members’ perceptions of: (J) recent failures, (2) excessive difficulties in current decision-making, and (3) moral dilemmas (Janis, 1982, p. 244). Janis (1972, 1982) postulated that the existence of these antecedent con- ditions generated eight symptoms which were evident in the fiascoes studied and serve as the primary means of identifying the occurrence of groupthink. The symptoms are: (1) an illusion of invulnerability, (2) collective efforts to rationalize, (3) an unquestioned belief in the group’s inherent morality, (4) stereotyped views of enemy leaders as weak or stupid, (5) direct pressure on a member who argues against the group's stereotypes, (6) self-censorship of deviations from the group consensus, (7) a shared illusion of unanimity, and (8) the emergence of self-appointed mind guards to screen adverse in- formation (Janis, 1982, p. 244). The recent theoretical development produced a framework consisting of four categories of variables causally related as shown in Fig. 1. The con- ditions for the occurrence of groupthink are a highly cohesive group, augmented by structural and decision context dysfunctions. These conditions serve as stimuli for the eight symptoms which are conducive to a high fre- quency of defects in decision-making. The decision-making defects subse- quently lead to lower quality decisions and lower performance. Given the popularity of the notion of groupthink, it is difficult to understand the dearth of empirical studies on the groupthink phenomenon. Three studies, Flowers (1977), Courtright (1978), and Leana (1985), attemp- ted to empirically validate the occurrence of groupthink in a laboratory set- ting. These studies do not provide strong primary support for the existence of groupthink, even though each researcher argued that he/she presented convincing secondary support for its presence. One major methodological problem with the Flowers (1977) and the Courtright (1978) studies was that neither was successful in producing the dominant antecedent condition posted by Janis (1972, 1982). Antecedent Conditions —___» Observable Outcomes -- cohesive groups -- structural faults symptoms» defects -» outcomes -- decision-making context Fig. 1. Groupthink theoretical frameworka2 Moorhead and Montanari Janis (1972, 1982) argues that a highly cohesive group is the primary antecedent of groupthink. However, both studies used various forms of ad hoc grouping and administered a posttreatment test for cohesiveness. It is very likely that the treatment itself created the cohesiveness measured/observ- ed after the fact. This is contrary to the causal sequence postulated by Janis which clearly posits cohesiveness as an antecedent, not a consequence. It could also help explain why Flowers’ results supported leader style and not group cohesiveness as the primary antecedent of groupthink. Perhaps a more serious error of previous groupthink research is one of omission. None of the studies cited attempted to test the entire theoretical framework. That is, no attempt was made to assess the presence of group- think symptoms nor did either study incorporate decision defects and deci- sion outcomes generated under crisis conditions. The most comprehensive empirical test to date (Courtright, 1978) included measures of only two of seven antecedent conditions, none of the symptoms, three of seven decision- making defects and outcomes measured on a subjective productivity rating instrument. As Manz and Sims (1982) conclude in their recent article on groupthink in autonomous work groups: “There is a need for more research on the potential threats posed by ‘groupthink’ to effective group decision- making in autonomous work groups (and the research could be based on a more rigorous quantitative base through the use of the groupthink symp- toms as behavioral categories for observational coding of behavior)” (p. 782). The purpose of this paper is to report the results of an initial study designed to empirically investigate the more comprehensive groupthink theoretical framework first presented by Janis in 1972 and expanded in 1982. RESEARCH METHOD Measures The initial phase of the study consisted of efforts to locate or develop scales to measure the groupthink variables and to empirically validate these scales. Results of the validation study are reported elsewhere (Moorhead & Montanari, 1982). After a thorough review of Janis’s (1972, 1982) work, a literature search was conducted to locate generally accepted, published scales to measure groupthink variables. Group cohesiveness was the only variable of which a published measure was available. Since Flowers (1977) and Cour- tright (1978) used confederates in a laboratory setting to induce/measure leadership style and decision defects, these techniques were inappropriate for the methodology used here.‘The Groupthink Phenomenon 403 Except for the cohesiveness measure, all scales were developed for the study. A summary of the measures developed follows. The antecedent con- ditions are (1) group cohesiveness (eight items), (2) insulation from outsiders (five items), and (3) promotional leadership practices (five items). The defects are (1) consideration of few alternatives (two items), (2) failure to reexamine preferred alternatives (two items), (3) failure to reexamine rejected alter- natives (two items), (4) rejection of opinions of experts (two items), (5) selec- tive use of new information (two items), and (6) absence of contingency plans (two items). The symptoms are (1) illusion of vulnerability (four items), (2) collective retionalization (two items), (3) inherent group morality (three items), (4) stereotyped view of others (three items), (5) direct group pressure on dis- sent (three items), (6) self-censorship of deviations (three items), (7) shared illusion of unanimity (three itmes), and (8) self-appointed mindguards (three items). All measures used 5-point, anchored Likert-type scales. Responses from 197 subjects were factor-analyzed to develop more valid and parsimonious scales. All items in each variable category, e.g., antece- dent conditions, were factor-analyzed initially. Separate factor analyses were performed for the symptoms and defects categories. Since Janis (1982) is unclear about the relationships among variables within a groupthink category and implies that they are independent, factor analysis with orthogonal rota- tion was used in all cases. Only those items with high and unambiguous loadings within a variable category were used in subsequent analyses. The factor analysis for the antecedents category produced three fac- tors with eigenvalues greater than 1.419 that explained 59.6% of the variance. Five of the original eight items loaded on the cohesiveness factor. The re- maining two factors, insulation and leadership promotion, had three and two items, respectively, that displayed high, unambiguous loadings. The antecedent variables conformed nicely to the factors suggested by Janis. Analysis of the items developed for the groupthink symptoms proceeded in a similar fashion. It resulted in only four clearly distinct factors which were not exactly consistent with the symptoms proposed by Janis. Items from several factors combined in the final factor structure as shown in Table I. The first factor, invulnerability, consisted of items that were concerned with the illusion of invulnerability and a negative view of outsiders. Rather than being conceptually distinct factors as suggested by Janis, the analysis indicated that the concepts inherent in them were very similar. Therefore, they were combined into one factor. The second factor, group morality, was made up of items concerned with the inherent morality of the group, a feeling of unanimity among the group, and rationalization attempts within the group. The self-censorship factor consisted only of items that were concerned with the self-censorship of discussion by members. The final symptom factor, discouraged dissent, was comprised of items concerned with the pressure put404 Moorhead and Montanari Table I. Final Factor Structure for Symptoms and Decision-Making, Defects Scales Final factor Number Primary topic structure of items of items ‘Symptoms Invulnerability 4 Illusion of invulnerability 2 Negative view of outsiders Group morality 2 Inherent morality of group 2 Unanimity and rationalization Self-censorship 2 Self-censorship Discouraged dissent 2 Pressure on dissenters 1 Negative view of dissenters Decision-making defects Few alternatives 2 Number of alternatives considered 1 No consideration of contingencies Lack of expert advice 2 Rejecting expert advice 1 Selective use of new information on dissenters and a negative view of dissenters. These findings suggest that the symptoms proposed by Janis may represent concepts that are so similar that a group member may not be able to distinguish between them. Therefore, the empirically-derived factors were used in all further analyses. Ina similar manner, the items from the decision-making defects scale were factor-analyzed, resulting in the factors shown in Table I. The first defect factor, few alternatives, consisted of items that were concerned with the number of alternatives considered and lack of consideration of contingen- cies. The second factor, lack of expert advice, consisted of items that were concerned with the rejection of expert advice and the selective use of new information. Note that Janis split the consideration of few alternatives into failure to reexamine preferred alternatives and failure to reexamine rejected alternatives as well as the consideration of few alternatives. Thus, the empirically-derived factors represent combinations of the concepts in the original defects proposed by Janis, similar to those combinations in the symp- toms scale. The analyses discussed above resulted in three antecedent variables, four symptom variables, and two defect variables with high construct validity that‘The Groupthink Phenomenon 405 Table II. Means, Standard Deviations, and Relability Coefficient for Groupthink Variables Variable Mean SD___ Reliability (a) Antecedent conditions Cohesion 4.27 0.58 7 Insulation 2.83 0.76 58 Leadership 3.72 0.72 48 ‘Symptoms Invulnerability 3.06 0.83 7 Morality 3.55 0.73 70 Self-censorship 2.22 0.86 67 Discouraged dissent 3.38 0.77 5S Defects in decision making Few alternatives Lack of expert advice 0.66 n 0.82 70 were used in subsequent analyses. Internal reliabilities were calculated for all empirically-derived scales. All alpha values were between 0.55-0.77 ex- cept for the leadership promotion antecedent variable (a = 0.48). These reliability coefficients and the mean and standard deviation for each variable are shown in Table IT. The remaining antecedent conditions were produced by selection of the experimental conditions. For example, the “lack of norms requiring methodical procedures” was frustrated for all groups by the imposition of severe time constraints on decision task accomplishment. “Homogeneity of backgrounds” was assured by sample selection. Finally, “high stress from external threats” was an integral part of the decision exercise used. The outcome measures used were group performance levels on two structured, timed decision tasks. Subjects were directed to evaluate data pro- vided and make decisions which would dramatically affect the profitability of their firm or their own survival. These exercises were selected because they simulated “high stake” decisions made under severe time constraints. These two conditions provided the “Provocative Situational Context” referred to by Janis as a necessary prerequisite for groupthink. Subjects Forty-five teams of 3-5 subjects were involved in this study. All sub- jects were senior level students enrolled in a Business Policies class at a large university in the Southwest. All student teams had participated in a highly competitive management simulation exercise for approximately 3 months prior to the research. Each team was in competition with all other teams in the class over the entire duration of the simulation exercise. According to406 Moorhead and Montanari Janis, this highly competitive environment involving relatively high stakes (course grades) should create highly cohesive task groups. All groups then participated in one of the two decision-making exercises. Team results for the exercise were recorded and members were required to individually com- plete the groupthink instrument. Finally, individual team member responses were aggregated to obtain a team score on the groupthink variables. Analysis Technique The level of analysis used in this study was the decision-making group. Since Janis (1972, 1982) describes groupthink as a group phenomenon, this and previous research have consistently used the group level of analysis. However, contrary to previous efforts, this study proposed to investigate the implied causal sequence of the Janis framework, rather than a few of its prin- ciple concepts. Therefore, path analysis was determined to be a more ap- propriate technique than ANOVA or MANOVA, given the cross-sectional nature of these data. RESULTS In path diagrams, one-way arrows from each determining variable to each variable dependent on it are used to represent the paths. Regression equations are developed by regressing each variable on all variables which precede it in the model. The numerical values assigned to each path are stan- dardized beta coefficients obtained from the regression equations. For paths where the standardized beta coefficients are small and not statistically signifi- cant, the paths are eliminated and the beta coefficients are recalculated. The path diagram for the groupthink model is shown in Fig. 2. These results show only one variable with significant effects on group performance: the antecedent condition, insulation. That is, groups that felt themselves to be most insulated from outsiders had lower performance. No other antece- dent condition, groupthink symptom, or decision-making defect had signifi- cant effects on group performance. The antecedent conditions had significant impacts on the groupthink symptoms and on decision-making defects. Cohesion, which Janis contends is the most important antecedent condition had a negative impact on self- censorship, a positive impact on dissent, and a negative impact on the defect, few alternatives. These results suggest that where groups were highly cohesive, there was less self-consorship, more discouragement of dissent, and more generation and evaluation of alternatives.‘The Groupthink Phenomenon 407 [sesorsnsp | se Fig. 2. Path analysis results with group performance ratings. Insulation, in addition to its negative impact on group performance, had a negative impact on invulnerability, a negative impact on lack of ex- pert advice, and a positive impact on alternative generation and evaluation. This suggests that insulation of the group decreases the feeling of in- vulnerability, increases the use of experts, and leads to fewer alternatives being generated and/or evaluated. Leadership practices had a positive impact on morality, a positive im- pact on discouraging dissent, and a negative impact on few alternatives. This suggests that the leadership practices in the group may lead to a feeling of morality, may lead to more discouragement of dissent with the group, and may lead to more alternatives being generated and/or evaluated. It is interesting to note that the decision-making defects, lack of expert advice and few alternatives generated and/or evaluated, have distinctly dif- ferent patterns of relationships with the antecedent conditions. The lack of expert advice is affected primarily by the three symptoms, self-censorship, discouraging dissent, and invulnerability, and is only directly impacted by one antecedent condition, insulation. The other two antecedent conditions, cohesion and leadership, impacted lack of expert advice only indirectly through self-censorship, discouraging dissent, and invulnerability. On the other hand, the few alternatives defect is impacted directly by all three antece- dent conditions, cohesion, insulatin, and leadership, with no direct effects from any of the groupthink symptoms. Several relationships strongly support the framework proposed by Janis. An increasing feeling of cohesiveness was accompanied by an increasing feel-408 Moorhead and Montanari ing that dissent was discouraged. Also, more promotion of a favored solu- tion by the leader was accompanied by increasing feelings of morality and discouragement of dissent. Finally, groups that felt that they were insulated had lower performance ratings. On the other hand, the empirically derived model suggests that several linkages were opposite to those predicted by the Janis framework. The most common were the antecedents of the feeling that there was no expert advice. This may indicate a facet of the model that needs further conceptual develop- ment. However, these may have been artifacts of the experimental situation. In the decision-making exercise, the groups may have felt confined by nor- mal classroom protocol to remain in the classroom for the scheduled class period, and thus did not feel as if they could seek outside counsel. Although they were not told that they could leave the room, not a single group asked to leave the room. It is also interesting to note that the decision-making defect, examina- tion of few alternatives, had no symptoms as direct antecedents. The three antecedents were the group characteristics, and two of these relationships were inverse to that predicted in the model—cohesiveness and leadership. An increasing feeling of cohesiveness and leadership promotion of prefer- red solutions were accompanied by examination of more alternatives. This may have resulted from the characteristics of this particular decision which had an unlimited number of possible solutions. Regardless of the feelings of the group and the actions of the leader, the group could still explore numerous courses of action. In summary, while these data do not show powerful support for Janis’s groupthink hypothesis, there are significant relationships among the antece- dent conditions, the groupthink symptoms, and the decision-making defects to warrant further discussion and investigation. In addition, these results need to be discussed in relation to the previous tests of groupthink previously reported. DISCUSSION The purpose of this research was to test the entire groupthink theoretical framework first presented by Janis in 1972 and expanded in 1982. Where other research efforts have used ad hoc groups brought together for the first time where researchers tried to induce cohesion quickly, this study utilized groups which had been together for 3 months interacting on highly signifi- cant decisions. Janis suggests that cohesiveness is a primary antecedent con- dition which means that it must develop over time and be present to some degree before the decision situation in which groupthink might occur. In this‘The Groupthink Phenomenon 409 study, all three antecedent conditions were truly “antecedent” since the groups had been together for 3 months. Other efforts to examine the groupthink hypothesis have not examin- ed the entire framework as posited by Janis. In this study, all four levels of concern to Janis were examined: the antecedent conditions, the group- think symptoms, decision-making defects, and group performance. Path analysis procedures were used to examine the direct and indirect effects of each variable on the others which follow it in Janis’s theoretical framework. These results indicate that insulation of the group most strongly affects group performance. This is in contrast to Flowers’ (1977) results supporting the leadership practices as leading to groupthink. However, these results show all three antecedent conditions affecting groupthink symptoms and decision- making defects. Although it was somewhat disappointing that there were no direct effects of symptoms or defects on group performance, we are encourag- ed by the relationships among antecedent conditions, symptoms, and defects. CONCLUSION ‘These findings show support for several key linkages in Janis’ group- think hypothesis and for the research methods used in this study to examine it. We have tried to examine the entire framework posited by Janis and over- come limitations of other research efforts. The relationships among antece- dent conditions, groupthink symptoms, and decision-making defects add to our understanding of how groupthink occurs, although their effects on group performance were not encouraging. The use of permanent, interacting groups and path analytic techniques have enabled us to test the entire framework suggested by Janis. The singular effect on group performance of insulation of the group may indicate that there are other variables which may be in- volved in the development of groupthink. REFERENCES CARTWRIGHT, D. Risk taking by individuals and groups: An assessment of research employing choice dilemmas. Journal of Personality and Social Psychology, 1971, 20, 361-378. COURTRIGHT, J. A. A laboratory investigation of groupthink. Communications Monographs, 1978, 45, 229-246, FLOWERS, M. L. A laboratory test of some implications of Janis's Groupthink Hypothesis. Journal of Personality and Social Psychology, 1977, 35, 888-896. JANIS, I. L. Victims of groupthink. Boston: Houghton Mifflin, 1972. JANIS, I. L. Groupthink Qnd ed.). Boston: Houghton Mifflin, 1982. | LEANA, C. R. A partial test of Janis’ Groupthink Model: Effects of group cohesiveness and leader behavior on defective decision making. Journal of Management, 1985, 11, 5-17.410 Moorhead and Montanari MANZ, C. C., & SIMS, H. P., The potential for “Groupthink” in autonomous work groups. ‘Human Relations, 1982, 35, 773-784. MOORHEAD, G., & MONTANARI, J. R. Groupthink: A research methodology. Proceedings of the 14th Annual Meeting of the American Institute for Decision Sciences, San Fran- cisco, 1982, pp. 380-382. VROOM, V. H., GRANT, L. D., & COTTON, T. S. The consequences of social interaction in group problem solving. Organizational Behavior and Human Performance, 1969, 4, 71-95. BIOGRAPHICAL NOTES GREGORY MOORHEAD is Associate Professor of Management at Arizona State University. He attended Texas Tech University and the University of Houston where he received a BS in Industrial Engineering and an MBA and PhD in Organizational Behavior and Management. His research interests include group decision making and integrative analysis of organization, group, job, and person relationships. During the 1984-1985 academic year, Professor Moorhead. ‘was on sabbatical leave from ASU and was a visiting Associate Professor of Management at Texas A & M University, JOHN R. MONTANARI is Associate Professor of Management at Arizona State University. His research interests include organizational structure and employee performance, employee decision behavior and performance, measures of organization effectiveness, and the strategic decision-making process. He received a DBA in organizational behavior and administrative policy from the University of Colorado, an MBA from the University of New Mexico, and a BS in Mechanical Engineering from the University of Dayton,