Somebody in your last difficult meeting used the word groupthink. It is one of the few pieces of social psychology that reached ordinary business vocabulary intact, and its evidentiary standing is considerably weaker than its fame.
Key Takeaway
Janis derived groupthink from case studies of foreign policy fiascoes including Pearl Harbor, the Bay of Pigs and Vietnam[1][2]. A reference work states that despite the popular appeal of the concept, empirical research evidence provides only equivocal support for the groupthink model, and that case studies of decision making by intact groups frequently find inconsistent evidence for the complete model[3]. A competing account, the ubiquity model, holds that the antecedent conditions are not necessary, with evidence that the symptoms have arisen without them[4].
Our Grades For These Claims
Applying the scheme from the first article in this series.
That group pressure can alter individual behaviour is Grade A. A reference work notes that Janis's basic assumption that individual behavior can be altered by group pressure is supported by a substantial body of empirical research[5]. Nothing in this article disputes that.
Janis's specific antecedent model is Grade C. Empirical support is described as equivocal, case studies find inconsistent evidence for the complete model, and a competing account holds the antecedents are unnecessary.
The claim that cohesion causes decision failures is Grade D as derived, for a methodological reason set out below that has nothing to do with whether it happens to be true.
Our own position, stated up front: this is the weakest evidentiary position of anything this series has covered, and it is the best known idea in the field. Those two facts are related.
A Note On Method
Everything here is verified to August 2026.
We did not obtain Janis's book and describe the theory only as reference works and later literature describe it.
We did not obtain any of the 1998 review papers that reassessed the theory at twenty-five years. We have their titles, authors, journal, volume and page ranges from a bibliographic index, and we report nothing about their findings that does not come from elsewhere.
Our sources are predominantly encyclopedia entries and reference works rather than primary research, which is a genuine limitation and one we flag at each use.
The central methodological argument in this article is ours, not something a source states, and the simulation illustrating it uses parameters we chose.
This article reviews research methods and organisational research. It is not management or governance advice.
The Theory
What Janis proposed.
Irving Janis, a Yale University psychologist, proposed that highly cohesive groups are likely to suffer from groupthink, a strong concurrence-seeking tendency that suppresses critical inquiry and results in faulty decision-making processes and flawed outcomes[1][6].
A reference work describes the phenomenon as a mode of thinking in which individual members of small cohesive groups tend to accept a viewpoint or conclusion that represents a perceived group consensus, whether or not the group members believe it to be valid, correct, or optimal[2].
Another puts the mechanism in terms of a process through which a group reaches a hasty or premature consensus and then becomes closed to outside ideas or alternative thoughts within, where the effort to achieve unanimity and agreement overrides the desire to consider a full range of policy alternatives[5].
Two observations, ours.
Janis chose the name deliberately. A source records that he selected the term because of its frankly Orwellian connotation, similar to doublethink and crimethink[1]. That is a rhetorical decision as much as a scientific one, and it matters for how the idea travelled.
And the theory first appeared in a 1971 issue of Psychology Today, before the 1972 book[1][2]. It entered public circulation through a popular magazine rather than a journal.
The Model's Structure
The formal apparatus, because the criticism turns on it.
Janis presented three categories of antecedents. First, moderate to high group cohesion is a necessary but not sufficient condition for groupthink. Structural faults and a provocative situational context are secondary antecedents[1].
A source lists the structural conditions as relating to the cohesiveness of a given decision-making group, the formal rules governing its decision-making process, the character of its leadership, the social homogeneity of participants, and the situational context they face[2].
Another summarises the required conditions as high cohesiveness within the group, isolation of the members from contrary views, and a lack of impartial leadership, with consequences including the illusion of invulnerability and unanimity and self-censorship, often leading to defective decision-making[4].
Two observations, ours.
The structure is antecedents produce symptoms produce bad outcomes. It is a causal chain, and each link is a separate empirical claim.
And cohesion is doing the foundational work. It is stated as necessary. If cohesion turns out not to be necessary, the model as specified is wrong, whatever else survives.
The Cases It Was Built From
Where the evidence came from, and this is the whole argument.
Janis examined a number of "fiascoes" in American foreign policy, and a source lists his interest as arising from wondering why people in authority frequently act in ways that are counterproductive[6].
The cases: the failure to anticipate the Japanese attack on Pearl Harbor in 1941, the Bay of Pigs invasion in 1961, and the prosecution of the Vietnam War from 1964 to 1967[6], together with the escalation of the Korean War and Great Britain's appeasement of Nazi Germany[3].
His conclusion, as the same source records: in each of these cases, the decisions occurred largely because of groupthink, which prevented contradictory views from being expressed and evaluated, and a group is especially vulnerable when its members are similar in background, when the group is insulated from outside opinions, and when there are no clear rules for decision making[6].
Read that sequence again and notice its shape. He started with outcomes he already knew were bad, looked for what they had in common, and named the common features as causes.
The Problem With That
The methodological objection. This section is our own argument and is not drawn from any source we obtained.
Selecting cases on the outcome you are trying to explain is a known and serious problem, and it has a specific consequence here.
If you examine only failures, you can establish what failures have in common. You cannot establish that those common features caused the failures, because you have no idea how often the same features accompany success.
Three consequences.
Any feature that is simply common in high-stakes decision groups will appear in most fiascoes, whether or not it does anything. Senior people who know each other well, working under time pressure, on a matter they cannot publicly discuss, is an accurate description of a great many important decisions, successful and otherwise.
So the observation that fiascoes involve cohesive insulated groups is compatible with cohesion causing failure and equally compatible with cohesion doing nothing at all.
And distinguishing those requires a comparison group: the rate of failure among cohesive groups against the rate among non-cohesive ones. That is not what the original method supplied.
A Simulation That Makes It Concrete
Our own illustration, with parameters we chose to demonstrate a mechanism rather than to estimate anything real.
Assume a thousand high-stakes decisions. Assume 60 percent are made by cohesive groups, because senior decision-making bodies tend to be cohesive. Assume 30 percent of decisions fail. And crucially, assume cohesion has absolutely no effect on whether a decision fails.
The resulting table is arithmetic. Cohesive and failed: 180. Cohesive and succeeded: 420. Not cohesive and failed: 120. Not cohesive and succeeded: 280.
Now do what the original method did, and look only at the failures.
Of 300 failures, 180 involved cohesive groups. That is 60 percent.
Two observations, ours.
A clear majority of fiascoes involve cohesive groups, and cohesion was set to have exactly zero effect. The 60 percent reflects how common cohesion is, not what it does.
Which means an investigator examining only fiascoes would find cohesion present in most of them and could reasonably conclude it mattered. The method produces that conclusion whether or not it is true.
We are not claiming cohesion has no effect. We are claiming that this method could not have told anyone either way, and that is a different and more damaging criticism than saying the conclusion was wrong.
The One Contrast Case
The partial answer Janis did supply, which deserves acknowledging.
A source records that Janis contrasted the fiascoes with what he calls vigilant appraisal, in which decision makers do rigorously and thoroughly consider all possible options, and holds up the Cuban missile crisis of October 1962 as the example[5].
Two observations, ours.
That is a real methodological move and it is to his credit. Including a successful case is exactly the right instinct, and many case-study literatures never do it.
But one positive case against five or six negative ones is not a comparison group. It cannot support a rate, and a rate is what the causal claim requires. It also carries an obvious risk: the successful case can be described in the theory's favourable terms as easily as the failures were described in its unfavourable ones, because the coder already knows how each turned out.
What The Empirical Research Found
What happened when the theory was tested rather than illustrated.
A reference work states the position: despite the popular appeal of the concept, empirical research evidence provides only equivocal support for the groupthink model, and case studies of decision making by intact groups frequently find inconsistent evidence for the complete model[3].
Another notes that the theory has been heavily criticized in recent years[5].
Two observations, ours.
The phrase the complete model is doing precise work. It suggests that individual pieces find support while the full antecedent-symptom-outcome chain does not, which is the ordinary fate of an elaborate model built from cases.
And we should be candid about our sourcing here. This is a summary judgment in a reference work rather than a meta-analysis we obtained. It is the consensus characterisation, reported by more than one such work, and it is not the same as a quantitative synthesis. We did not locate one.
The Ubiquity Model
The most damaging single finding, and it targets the foundation.
A source describes a competing account: according to the so-called ubiquity model of groupthink, the antecedent conditions are not necessary for the groupthink to occur. In fact, there is evidence that the symptoms have arisen without the presence of such antecedents[4].
We did not obtain the paper proposing this model and report only this characterisation.
Three consequences, ours.
Recall that Janis specified cohesion as necessary but not sufficient[1]. A necessary condition is one without which the phenomenon cannot occur.
If the symptoms arise without the antecedents, then the antecedents are not necessary, and the model's central structural claim fails as stated.
And it would mean the phenomenon, if real, is more general and less diagnosable than the theory holds. You could not identify at-risk groups by looking for cohesion and insulation, because groups lacking both would show it too.
Why That Finding Is Decisive
What the ubiquity result does to the practical value of the theory. This section is our own reasoning.
The whole managerial appeal of groupthink is diagnostic. It tells you which groups to worry about: the tight-knit, insulated, homogeneous ones with a directive leader.
Two consequences if the antecedents are unnecessary.
The diagnostic value collapses. If the symptoms can appear anywhere, then knowing your group is cohesive tells you little, and knowing it is not tells you nothing reassuring.
And a manager who has deliberately broken up the antecedents, by rotating membership, bringing in outsiders and adopting formal procedures, has no assurance from this literature that the underlying problem has gone.
That is a different situation from a theory being merely small or contested. A model whose specified conditions turn out not to be required has lost the thing it was being used for.
An Entire Journal Issue
The scale of the reassessment, which is itself informative.
A bibliographic index records that Organizational Behavior and Human Decision Processes, volume 73, issue 2-3, published in February 1998, carried a series of papers on the theory at twenty-five years[7].
Among them: Esser, Alive and Well after 25 Years: A Review of Groupthink Research, at pages 116–141. Kramer, Revisiting the Bay of Pigs and Vietnam Decisions 25 Years Later: How Well Has the Groupthink Hypothesis Stood the Test of Time?, at 236–271. Peterson, Owens, Tetlock, Fan and Martorana, Group Dynamics in Top Management Teams: Groupthink, Vigilance, and Alternative Models of Organizational Failure and Success, at 272–305. Whyte, Recasting Janis's Groupthink Model: The Key Role of Collective Efficacy in Decision Fiascoes, at 185–209. And further papers by McCauley, 't Hart, Moorhead and colleagues, and Turner and Pratkanis[7].
We obtained none of these papers and report nothing about their contents.
Two observations we think the titles alone support, and we mark them as inference.
A field that devotes an entire journal issue to whether an idea has held up is a field with doubts.
And the recurring words in those titles, being revisiting, recasting, alternative models and how well has it stood the test of time, are the vocabulary of reassessment. That is weaker evidence than reading the papers, and we present it as such.
What Does Survive
The part that should not be thrown out.
A source is explicit: Janis's basic assumption that individual behavior can be altered by group pressure is supported by a substantial body of empirical research[5].
Another notes that groupthink has a close relation to the Solomon Asch line-matching paradigm, where members holding different opinions from the majority are under pressure to conform and suppress their true opinions[4].
Three observations, ours.
The underlying phenomenon of conformity pressure is not in dispute. What is disputed is Janis's specific model of when it produces bad decisions and how to spot it.
That distinction matters commercially, because a manager can accept the first entirely while declining to use the second as a diagnostic.
And the second article in this series described the same phenomenon from a better-evidenced direction. Whether it is safe to voice a dissenting view is what psychological safety measures, and that literature carries a meta-analysis and a systematic review behind it.
Why It Spread Anyway
Our own account of the theory's success, offered as reasoning.
Four features made groupthink unusually transmissible.
The name. Chosen for its Orwellian connotation[1], it is memorable, slightly sinister and immediately intelligible.
The cases. Pearl Harbor and the Bay of Pigs are stories everybody knows, which makes the theory feel demonstrated rather than argued.
Publication route. It appeared in a popular magazine first[1], reaching a general audience before it reached its critics.
And it explains failure without blaming anyone in particular, which is an enormously comfortable property for an organisation reviewing a bad decision.
A source records that the concept quickly gained remarkably broad and firm acceptance, dominating the literature on group decision making for decades[1]. This series has now documented that pattern in brainstorming, in the feedback sandwich, in the jam study and in growth mindset. Memorability and evidentiary weight are close to uncorrelated, and the business world selects on the first.
The Diagnosis Nobody Can Refute
A practical problem with the term in ordinary use. This section is our own.
Consider what happens when someone says a meeting suffered from groupthink.
If the group agreed, that is concurrence-seeking. If the group agreed after discussion, the discussion was insufficiently critical. If someone dissented and was persuaded, that is self-censorship or conformity pressure. If nobody dissented, that is the illusion of unanimity.
Three consequences.
As deployed in ordinary business conversation, the term is close to unfalsifiable. There is no observable meeting outcome that counts against it.
Which means it functions less as a diagnosis than as a way of expressing disagreement with a decision in language that sounds diagnostic.
And it is almost always applied retrospectively, after the outcome is known, which reproduces at the level of a single meeting the same selection problem the original theory had at the level of history.
The Growing List Of Cases
An observation about how the theory has been applied since.
A reference work lists subsequent applications: the WorldCom accounting fraud, Nazi Germany's decision to invade the Soviet Union in 1941, Ford's decision to market the Edsel, Gruenenthal Chemie's decision to market thalidomide, the tragedy at Kent State, the Challenger and Columbia space shuttle disasters, a European Union sanctions decision, and a city council's decision to ignore earthquake preparation procedures[3]. Others add Watergate[5], the interpretation of intelligence on weapons of mass destruction before the Iraq War[2], and the grounding of Swissair and, to some degree, Brexit[4].
Two observations, ours.
Every single item on that list is a known disaster. The theory has been applied for fifty years by continuing to select on the outcome, which means the accumulating case list is not accumulating evidence, however long it grows.
And the range is now so wide, from a shuttle launch to a pharmaceutical approval to a referendum, that a theory fitting all of them is doing so at a level of generality that constrains very little.
The Remedies Might Be Fine Anyway
An important qualification, and it is ours.
A source notes that Janis suggested several methods to prevent groupthink[1], and the familiar list includes assigning a devil's advocate, bringing in outside experts, having the leader withhold their view initially, and splitting into subgroups.
Three observations.
A remedy can be good practice even if the theory motivating it is wrong. Those procedures have independent justifications, and the tenth article in this series found direct evidence for one of them: separating independent generation from group discussion is what the Nominal Group Technique enforces, and the brainstorming literature supporting it is considerably stronger than the groupthink literature.
The leader-withholding practice connects to the fifth article, on anchoring: a senior person's stated view is a number in the room before anyone else speaks.
And the cost of these procedures is low, which the ninth article argued is the relevant consideration when evidence is contested. A cheap procedure under a weak theory is a reasonable bet; an expensive programme under one is not.
Where This Sits Against Better Evidence
Placing this article among its neighbours. This section is our own.
This series has now covered three ways groups fail, and their evidentiary standing runs in the opposite direction to their fame.
Brainstorming productivity loss is the best evidenced: repeated experiments since 1958, a meta-analysis, and the mechanism isolated across four studies that eliminated the alternatives. Almost nobody outside the field has heard of production blocking.
Task and relationship conflict is middling: two meta-analyses that disagree on the main effect but converge on the moderator. Moderately known.
Groupthink is the weakest: equivocal support, inconsistent case evidence, a competing model that denies its antecedents are necessary. Universally known.
Our own conclusion: if a business wants to improve how its groups decide, the evidence is strongest for the least famous intervention, which is separating generation from discussion, and weakest for the one everybody already talks about.
What To Do
Stop using groupthink as a diagnosis. As deployed in ordinary conversation there is no meeting outcome that counts against it, which makes it a way of expressing disagreement rather than identifying a cause.
Distrust any theory built only from disasters. Examining failures tells you what failures have in common, not what caused them, because you never observe how often the same features accompany success.
Ask for the comparison group. The question is not whether cohesive groups fail, it is whether they fail more often than non-cohesive ones.
Do not assume breaking up the antecedents solves it. A competing model holds the symptoms arise without cohesion or insulation, which if right removes the diagnostic value entirely.
Keep the conformity finding. That group pressure alters individual behaviour is well supported and is not what this article disputes.
Prefer psychological safety as the construct. Whether it is safe to dissent is measurable, has a meta-analysis and a systematic review behind it, and is what you actually care about.
Run the cheap remedies anyway. Withholding the leader's view, splitting into subgroups and bringing in outsiders cost almost nothing and have independent support.
Separate generation from discussion. That is the intervention with the strongest evidence of anything in this territory, and it is the least discussed.
Notice the retrospective habit. Groupthink is almost always invoked after the outcome is known, which reproduces the original method's flaw at the scale of a single meeting.
The Limits Of This Analysis
Several caveats matter. This article reviews research methods and organisational research and is not management or governance advice. Everything is verified to August 2026. We did not obtain Janis's book and describe the theory only as reference works and later literature describe it. We obtained none of the 1998 reassessment papers; we have their titles, authors, journal, volume and page ranges from a bibliographic index and report nothing about their contents, and our reading of the pattern in their titles is explicitly marked as inference from titles alone. We did not obtain the paper proposing the ubiquity model and report only a characterisation of it from a working paper. Our sources are predominantly encyclopedia entries and reference works rather than primary research, and the summary judgment that empirical support is equivocal is a reference work's characterisation rather than a quantitative synthesis, which we did not locate. The central methodological argument in this article, that selecting cases on the outcome cannot establish causation, is our own and is not drawn from any source we obtained. The simulation is ours, uses parameters we chose to demonstrate a mechanism, estimates nothing about real decisions, and appears in no publication. The unfalsifiability argument, the account of the theory's transmissibility, the observation about the growing case list, and the comparison with better-evidenced literatures are our own reasoning. We are not asserting that cohesion has no effect on decision quality; we are asserting that the original method could not have established that it does.
Frequently Asked Questions
Is groupthink real?
What is wrong with the original evidence?
What is the ubiquity model?
Should we abandon the remedies?
What should we use instead?
Are you saying cohesion is harmless?
References
- SAGE Encyclopedia of Management Theory, entry on Groupthink, on Janis having proposed that highly cohesive groups are likely to suffer from groupthink, a strong concurrence-seeking tendency that suppresses critical inquiry and results in faulty decision-making processes and flawed outcomes; on his choosing the term because of its frankly Orwellian connotation, similar to doublethink and crimethink; on his discussing major historical fiascoes including Pearl Harbor, the escalation of war in Korea, the Bay of Pigs and Vietnam; on the concept first being presented in a 1971 issue of Psychology Today and quickly gaining remarkably broad and firm acceptance, dominating the literature on group decision making for decades; on Janis presenting three categories of antecedents, with moderate to high group cohesion a necessary but not sufficient condition and structural faults and a provocative situational context as secondary antecedents; and on his suggesting several methods to prevent groupthink. Note: a reference work, not primary research. sk.sagepub.com
- Encyclopaedia Britannica, entry on groupthink, on groupthink being a mode of thinking in which individual members of small cohesive groups tend to accept a viewpoint representing a perceived group consensus whether or not they believe it valid, correct or optimal; on the theory being developed by Irving Janis in his 1972 study Victims of Groupthink: A Psychological Study of Foreign-Policy Decisions and Fiascoes; on Janis identifying structural conditions relating to cohesiveness, the formal rules governing decision-making, the character of leadership, the social homogeneity of participants and the situational context; and on the notion being revived to help explain the interpretation of intelligence regarding weapons of mass destruction before the Iraq War. Note: an encyclopedia entry, not primary research. britannica.com
- SAGE encyclopedia entry on Groupthink in group processes, on groupthink being extreme concurrence-seeking predicted to result in highly defective decisions; on the concept having been developed to explain the Bay of Pigs, Pearl Harbor, Britain's appeasement of Nazi Germany, and escalation of the Vietnam and Korean Wars; on subsequent application to the WorldCom accounting fraud, the 1941 invasion of the Soviet Union, the Edsel, thalidomide, Kent State, the Challenger and Columbia disasters, an EU sanctions decision and a city council's earthquake preparation decision; and on empirical research evidence providing only equivocal support for the groupthink model, with case studies of decision making by intact groups frequently finding inconsistent evidence for the complete model. Note: a reference work. Its summary judgment on the evidence is a characterisation rather than a quantitative synthesis, and we did not locate one. sk.sagepub.com
- Working paper on group polarization models, on the theory of groupthink having been introduced by Janis in 1971 as a concurrence-seeking tendency; on its close relation to the Solomon Asch line-matching paradigm, with members holding minority opinions under pressure to conform and suppress their true opinions; on the required antecedent conditions including high cohesiveness, isolation from contrary views and a lack of impartial leadership; on consequences including the illusion of invulnerability and unanimity and self-censorship; on examples including the Bay of Pigs, Pearl Harbor, Korean and Vietnam escalation, the grounding of Swissair and, to some degree, Brexit; and on the ubiquity model holding that the antecedent conditions are not necessary, with evidence that the symptoms have arisen without them. Note: a working paper. We did not obtain the paper proposing the ubiquity model and report only this characterisation. arxiv.org
- SAGE encyclopedia entry on Groupthink in international political science, on the theory having been heavily criticised in recent years while Janis's basic assumption that individual behavior can be altered by group pressure is supported by a substantial body of empirical research; on groupthink being defined as a process through which a group reaches a hasty or premature consensus and becomes closed to outside ideas, with the effort to achieve unanimity overriding the desire to consider a full range of alternatives; on high group cohesion or excessive camaraderie lying at the heart of the phenomenon; on Janis illustrating the theory with Pearl Harbor, the Bay of Pigs and the 1965 Vietnam escalation; on his application of it to Watergate and the Challenger launch decision; and on his contrasting these with vigilant appraisal, holding up the Cuban missile crisis of October 1962. Note: a reference work, not primary research. sk.sagepub.com
- SAGE Encyclopedia of War: Social Science Perspectives, entry on Groupthink, on Irving Janis being a Yale University psychologist who conducted the initial research; on his foundation in the American Soldier Project, a study of the effects of extreme stress on group cohesiveness; on his examining a number of fiascoes in American foreign policy including the failure to anticipate Pearl Harbor in 1941, the Bay of Pigs in 1961 and the prosecution of the Vietnam War from 1964 to 1967; on his wondering why people in authority frequently act in ways that are counterproductive; on his concluding that in each case the decisions occurred largely because of groupthink, which prevented contradictory views from being expressed and evaluated; and on a group being especially vulnerable when members are similar in background, when it is insulated from outside opinions, and when there are no clear rules for decision making. Note: a reference work, not primary research. sk.sagepub.com
- IDEAS/RePEc bibliographic index for Organizational Behavior and Human Decision Processes, volume 73, issue 2-3, February 1998, listing Esser, J. K., Alive and Well after 25 Years: A Review of Groupthink Research, 116–141; Kramer, R. M., Revisiting the Bay of Pigs and Vietnam Decisions 25 Years Later: How Well Has the Groupthink Hypothesis Stood the Test of Time?, 236–271; Peterson, R. S., Owens, P. D., Tetlock, P. E., Fan, E. T., & Martorana, P., Group Dynamics in Top Management Teams: Groupthink, Vigilance, and Alternative Models of Organizational Failure and Success, 272–305; Whyte, Recasting Janis's Groupthink Model: The Key Role of Collective Efficacy in Decision Fiascoes, 185–209; McCauley, C., Group Dynamics in Janis's Theory of Groupthink: Backward and Forward, 142–162; 't Hart, P., Preventing Groupthink Revisited, 306–326; Moorhead, G., Neck, C. P., & West, M. S., The Tendency toward Defective Decision Making within Self-Managing Teams, 327–351; and Turner, M. E., & Pratkanis, A. R. (1998). Note: a bibliographic index only. We obtained none of these papers and report nothing about their contents; our reading of the pattern in their titles is inference from titles alone. ideas.repec.org
This article reviews research methods and organisational research and is not management or governance advice. Janis's book was not obtained. None of the 1998 reassessment papers were obtained and nothing is reported about their contents. Sources are predominantly reference works rather than primary research. The central methodological argument and the simulation illustrating it are the authors' own, appear in no publication, and demonstrate a mechanism rather than estimating anything about real decisions. This article does not assert that group cohesion is harmless.