BUILD A BET TER

PROCESS

ORIGINALLY PUBLISHED DECEMBER 2014

Making
Dumb Groups
Smarter
The new science of group
decision-making
→ by CASS R. SUNSTEIN and REID HASTIE

S I NC E T H E B EG IN N IN G of human history, people have made

decisions in groups. As the saying goes, two heads are better
than one. If so, then three heads should be better than two,
and four better still. With a hundred or a thousand, then,
things are bound to go well—hence the supposed wisdom
of crowds.
The advantage of a group, wrote one early advocate of
collective intelligence—Aristotle—is that “when there are
many who contribute to the process of deliberation, each
can bring his share of goodness and moral prudence…some
appreciate one part, some another, and all together appreciate
all.” The key is information aggregation: Different people take

76 HBR Special Issue

Fall 2023 Illustration by ANDREA D’AQUINO
HBR Special Issue
Fall 2023 77
 BUILD A BET TER PROCESS
MAKING DUMB GROUPS SMARTER

note of different “parts,” and if those teams. But little of this work has trickled • They focus on what everybody knows
parts are properly aggregated, they into the public consciousness, and it has already—and thus don’t take into ac-
will lead the group to know more (and yet to have a noticeable effect on actual count critical information that only one
better) than any individual. practice. It’s time for that to change. or a few people have.
Unfortunately, groups all too often We aim to bring behavioral research
fail to live up to this potential. Compa- into direct contact with the question
nies bet on products that are doomed to of group performance—to describe the Amplifying Errors
fail, miss out on spectacular opportuni- main ways in which groups go astray With the psychologists Daniel
ties, pursue unsuccessful competitive and to offer some simple suggestions Kahneman and the late Amos Tversky
strategies. In governments, policy judg- for improvement. in the vanguard, behavioral scientists
ments misfire, hurting thousands or have identified some common mental
even millions of people in the process. shortcuts (known as heuristics) and
“Groupthink” is the term most often Why Do Errors Occur? biases that lead individuals astray. The
applied to the tendency of groups to go Groups err for two main reasons. The planning fallacy, for example, leads
astray. Popularized in the early 1970s first involves informational signals. us to underestimate how much time
by the psychologist Irving Janis, it has Naturally enough, people learn from projects will take and how much money
deservedly entered the popular lexicon. one another; the problem is that groups they’ll cost. Overconfidence leads us
But Janis’s contribution is more an often go wrong when some members to believe that our forecasts are more
evocative narrative than either a scien- receive incorrect signals from other accurate and precise than in fact they
tific account of how groups go wrong members. The second involves reputa- are. The availability heuristic leads
or helpful guidance for group success. tional pressures, which lead people to si- us to seize on whatever springs most
Many researchers have tried to find lence themselves or change their views readily to mind, because it is memora-
experimental evidence to support his in order to avoid some penalty—often, ble or we recently experienced it. The
specific claims about how cohesion and merely the disapproval of others. But if representativeness heuristic leads us to
leadership styles shape group behavior, those others have special authority or believe that things or events or people
to little avail. wield power, their disapproval can pro- that are similar in one way are similar
Since Janis produced his theory, duce serious personal consequences. in other ways, too. Egocentric bias leads
though, psychologists and other behav- As a result of informational signals us to exaggerate the extent to which
ioral scientists have built up a rich base and reputational pressures, groups run our tastes and preferences are typical.
of evidence on how and when individ- into four separate though interrelated The sunk-cost fallacy leads us to stick
ual decision-makers blunder. This work problems. When they make poor or with a hopeless project because we have
has attained scientific acclaim (includ- self-destructive decisions, one or more already invested so much in it. Framing
ing several Nobel prizes) and wide- of these problems are usually to blame: effects influence our decisions accord-
spread popularity thanks to best sellers • Groups do not merely fail to correct ing to the semantics of how the options
such as Daniel Kahneman’s Thinking, the errors of their members; they am- are presented. (For example, people
Fast and Slow, Dan Ariely’s Predictably plify them. are more likely to agree to an operation
Irrational, and Nudge (which one of us, • They fall victim to cascade effects, as if they are told that 90% of people are
Sunstein, coauthored with the econo- group members follow the statements alive after five years than if they are
mist Richard Thaler). and actions of those who spoke or acted told that 10% of people are dead after
A smaller but nonetheless substantial first. five years.)
body of research—some of it our own— • They become polarized, taking up For our purposes, the central ques-
has focused on the decision-making positions more extreme than those they tion is whether groups can avoid or
strengths and weaknesses of groups and held before deliberations. mitigate these errors. Experimental

78 HBR Special Issue

Fall 2023
 Many groups end up thinking that their
ultimate convergence on a shared view was inevitable.
Beware of that thought.

evidence indicates that they usually do better than individuals at overcoming

not. The psychologists Roger Buehler, egocentric bias. An individual will focus
Dale Griffin, and Johanna Peetz have on his own tastes—what he likes and
found, for example, that the planning what he doesn’t. If he consults with oth-
fallacy is aggravated in groups. That is, ers, he is likely to learn that his tastes
groups are even more optimistic than are idiosyncratic. In such cases, group
Idea in Brief
individuals when estimating the time deliberation supplies an important
and resources necessary to complete a corrective. Note that we’re less apt to THE PROBLEM
task; they focus on simple, trouble-free get that corrective if the group consists Groups often fail to live up to
scenarios for their future endeavors. of like-minded people. The influence of their potential as decision-
Similarly, Hal R. Arkes and Catherine the availability heuristic, too, is slightly making bodies. Instead of
aggregating the knowledge and
Blumer have shown that groups are reduced in groups. Individual members
wisdom of their members, they
even more likely than individuals to may each rely on “what comes to mind,”
end up making bigger errors
escalate their commitment to a course but they are likely to have different than individuals would.
of action that is failing—particularly memories, yielding a more representa-
if members identify strongly with the tive sample at the group level. WHY IT HAPPENS
group. There is a clue here about why The larger point, however, is that Group members take informa-
companies, states, and even nations of- many individual biases are not system- tional signals from what others
say, even when the information
ten continue with doomed projects and atically corrected at the group level
is wrong or misleading, while
plans. Groups have also been found to and often get worse. The mechanisms reputational pressures can
increase, rather than to lessen, reliance by which errors are compounded can cause them to silence them-
on the representativeness heuristic; to usually be found in the other three selves or change their views in
be more prone to overconfidence than problems of group decision-making. order to fit in. As a result, groups
individual members; and to be more often amplify individual errors,
influenced by framing effects. stampede toward bad decisions,
Both informational signals and rep- Cascading to the foment polarization and extrem-
ism, and ignore information that
utational pressures are at work here. If Wrong Answer isn’t already widely held.
most members of a group tend to make The human brain may be wired from
certain errors, then most people will see birth to synchronize with and imitate THE SOLUTION
others making the same errors. What other people. It is no exaggeration to say Leaders can structure group
they see serves as “proof” of erroneous that herding is a fundamental behavior deliberations to make them
more likely to succeed. One very
beliefs. Reputational pressures play a of human groups. When it comes to
simple way is to let others speak
complementary role: If most members group decisions and information flow,
first. Another is to assign spe-
of the group make errors, others may the favored term among social scien- cific roles or areas of expertise
make them simply to avoid seeming tists is “cascade”—a small trickle in one to members of the group. The
disagreeable or foolish. direction that soon becomes a flood. key is encouraging individuals
Fortunately, we have evidence that Consider a brilliant study of music to share their diverse knowledge
deliberating groups can correct or downloads by the sociologists Matthew rather than suppress it.
reduce certain biases. For problems Salganik, Peter Dodds, and Duncan
that have “eureka” solutions (the right Watts. They allowed test subjects to
answer, once announced, is clear to all), listen to and download one or more of
groups do well even if individual mem- 72 songs by new bands. In the control
bers start out biased. Groups are also group, individuals were told nothing

HBR Special Issue

Fall 2023 79
 BUILD A BET TER PROCESS
MAKING DUMB GROUPS SMARTER

A Tale of Two Cities even if it would have failed otherwise.

Many groups end up thinking that their
ultimate convergence on a shared view
was inevitable. Beware of that thought.
To examine the phenomenon ensure that the Boulder 2. Deliberation decreased The convergence may well be an artifact
of group polarization, the residents were in fact left the diversity of opinion
of who was the first to speak—and
two of us (along with the so- of center and the Colorado among group members.
cial scientist David Schkade) Springs residents were right Before the groups started hence of what we might call the archi-
created an experiment in of center. to deliberate, many of them tecture of the group’s discussions.
group deliberation—one Group members were showed considerable diver- Two kinds of cascades—informa-
that, we believe, accurately asked first to record their gence in individual opinions. tional and reputational—correspond
reflects much deliberation in views individually and anon- Discussion brought liberals
to our two main sources of group error.
the real world. ymously and then to delib- in line with one another and
We recruited citizens erate together in an effort to conservatives in line with In informational cascades, people
from two Colorado cities reach a group decision. After one another. After a brief silence themselves out of deference to
and assembled them in the deliberations the partic- period of discussion, group the information conveyed by others.
small groups (usually six ipants were again asked to members showed a lot less In reputational cascades, they silence
people), all from the same record their views individu- variation in the anonymous
themselves to avoid the opprobrium of
city. The groups were asked ally and anonymously. Here’s expression of their private
to deliberate on three of the what we found: views. others.
most contested issues of the Here’s an example of an informa-
time: climate change, affir- 1. People from Boulder 3. Deliberation sharply tional cascade in jury deliberations,
mative action, and same-sex became a lot more liberal, increased the disparities which has important implications
civil unions. The two cities and people from Colorado between the views of Boul-
for business. One of us (Hastie) has
were Boulder, known by Springs became a lot more der citizens and Colorado
its voting patterns to be conservative. Not only Springs citizens. Before conducted dozens of mock-jury studies
predominantly liberal, and were the group “verdicts” deliberation, many people’s with thousands of volunteer jurors,
Colorado Springs, known more extreme than the opinions overlapped many of them from big-city jury pools.
by its voting patterns to be pre-deliberation averages between the two cities. In these studies the volunteers privately
predominantly conserva- of group members, but the After deliberation, group
write down their preferred verdicts
tive. We did a reality check anonymous views of individ- dynamics left liberals and
on the participants before ual members became more conservatives much more before deliberations begin and indicate
the experiment started to extreme as well. sharply divided. how confident they are in their judg-
ments. Deliberations then start, as they
often do in real trials, with a straw vote
to see where everyone stands. The vote
about what others had downloaded or very top, and the best songs never at circles the jury table and frequently
liked and were left to make indepen- the very bottom, essentially anything begins with a small set of two or three
dent judgments. In other groups, the else could happen. If a song benefited jurors who favor, with increasing confi-
participants could see how many people from a burst of early downloads, it might dence, the same verdict.
had previously downloaded particular do quite well. Without that benefit, it In one mock trial, jurors 1, 2, and 3 en-
songs. The researchers were testing how might be a failure. And as the research- dorsed a verdict of second-degree mur-
much difference it made, in terms of ul- ers later found, these effects occurred der both privately and in the straw vote.
timate numbers of downloads, if people even if they lied to the test subjects about Juror 4 had voted not guilty and indi-
could see the behavior of others. which songs were downloaded a lot. cated the highest level of confidence
It made a huge difference. Although If a project, a product, a business, in his choice on the pre-deliberation
the worst songs (as established by the a politician, or a cause gets a lot of private ballot. What did juror 4 do
control group) never ended up at the support early on, it can win over a group when confronted with three second-

80 HBR Special Issue

Fall 2023
 The innovative thinking of individual members is
suppressed by the powerful ideas generated by other members.

degree murder verdicts? He paused for a as fashion. In both cases, groups lose ing Harry Potter?) does well, businesses
second and then said, “Second degree.” valuable information.) will react strongly, eagerly looking for a
Juror 7, an undecided vote, suddenly “Political correctness,” a term much proposal or a project that seems similar.
spoke up and asked, “Why second used by the political right in the 1990s, A by-product of availability is “asso-
degree?” The researchers saw a deer- is hardly limited to left-leaning aca- ciative blocking” or “collaborative fix-
in-the-headlights expression flit across demic institutions. In both business and ation,” whereby strong ideas block the
juror 4’s face before he replied, “Oh, it’s government there is often a clear sense recollection of other information. This
just obviously second degree.” We have that a certain point of view is the proper phenomenon is a big problem when
no doubt that similar scenarios play out one and that those who question or re- a group sets itself the task of generat-
every day in jury rooms, boardrooms, ject it, even for purposes of discussion, ing creative solutions. The innovative
and conference rooms all over the do so at their peril. They are viewed as thinking of individual members is sup-
world. “difficult,” “not part of the team,” or, in pressed by the powerful ideas generated
A reputational cascade has a extreme cases, as misfits. by other members.
different dynamic. Group members The group members in the examples In the actual world of group decision-
think they know what is right, but they above are, in a sense, entirely rational. making, of course, people may not know
nonetheless go along with the group They care about their reputations, but whether other members’ statements
in order to maintain the good opinion there’s nothing irrational about that. arise from independent information,
of others. Suppose, for example, that As noted, however, people use heuris- an informational cascade, reputational
Albert suggests that his company’s new tics, which can lead them astray, and pressures, or the availability heuristic.
project is likely to succeed. Barbara is are subject to biases. For the purposes They often overestimate the extent to
not confident that he’s right, but she of understanding how cascade effects which the views of others are based on
concurs because she wishes not to seem work, the most important heuristic independent information. Confident
ignorant, adversarial, or skeptical. If involves availability: A vivid idea or (but wrong) group decisions are a result.
Albert and Barbara seem to agree that example moves rapidly from one person
the project will go well, Cynthia not to another, eventually producing a
only won’t contradict them publicly widespread belief within a group and Polarizing Groups
but might even appear to share their possibly a city, a state, or even a nation. Polarization is a frequent pattern
judgment—not because she believes it In the area of risk, availability with deliberating groups. It has been
to be correct (she doesn’t), but because cascades are common. A particular found in hundreds of studies in more
she doesn’t want to face their hostility event—involving a dangerous pesticide, than a dozen countries. We found it in
or lose their good opinion. Once Albert, a hazardous waste dump, a nuclear dramatic form when we conducted an
Barbara, and Cynthia offer a united power accident, an act of terrorism— experiment in which residents of two
front on the issue, their colleague David may become well-known to the group, Colorado cities discussed their political
will be most reluctant to contradict even iconic. If so, it will alter members’ beliefs (see the sidebar “A Tale of Two
them, even if he’s pretty sure they’re perceptions of a process, a product, Cities”).
wrong and has excellent reasons for or an activity. Availability cascades The earliest experiments on the
that belief. (Incipient evidence indi- are familiar in business, too. Reports polarizing effects of deliberation
cates that women are especially likely of a success or a failure may spread like involved risk-taking behavior, with a
to self-censor during discussions of wildfire within or across companies, clear finding that people who are ini-
stereotypically male subjects, such leading to judgments about other, tially inclined to take risks become still
as sports, and that men are especially apparently similar events or products. If more so after they deliberate with one
likely to self-censor during discussions a movie (Star Wars?), a television show another. (Examples of risky decisions
of stereotypically female subjects, such (The Walking Dead?), or a book (involv- include accepting a new job, investing

HBR Special Issue

Fall 2023 81
 BUILD A BET TER PROCESS
MAKING DUMB GROUPS SMARTER

in a foreign country, escaping from a with David Schkade, Lisa Ellman, and nant position in order to preserve their
prisoner-of-war camp, and running for Andres Sawicki) has found that both self-presentation.
political office.) On the basis of this evi- Democratic and Republican appoin- The third reason stresses the close
dence the conventional wisdom became tees show far more ideological voting links among three factors: confidence,
that group deliberation produced a patterns when sitting with other judges extremism, and corroboration by
systematic “risky shift.” appointed by a president of the same others. When people lack confidence,
Later studies called this conclusion party. If you want to know how an they tend to be moderate. The great
into question—and created a puzzle. appellate judge will vote in an ideologi- American judge Learned Hand once
On many of the same issues on which cally contested case, you might want to said, “The spirit of liberty is the spirit
Americans made a risky shift, Taiwan- find out whether she was appointed by which is not too sure that it is right.” As
ese participants made a cautious shift. a Republican or a Democratic president. people gain confidence, they usually
Even among American participants, It’s a pretty good predictor. But in many become more extreme in their beliefs,
deliberation sometimes produced cau- areas of the law, an even better predictor because a significant moderating
tious shifts. Cautious shifts took place is who appointed the other judges on factor—their own uncertainty about
most often in decisions about whether the panel. whether they are right—has been
to marry and whether to board a plane Why does group polarization occur? eliminated. The agreement of others
despite severe abdominal pain. There are three principal reasons: tends to increase confidence and thus
What explains these unruly findings? The first and most important in- extremism.
As the psychologists Serge Moscovici volves informational signals—but
and Marisa Zavalloni discovered de- with a few twists. Group members pay
cades ago, members of a deliberating attention to the arguments made by Focusing on “What
group will move toward more-extreme other group members. Arguments in Everybody Knows”
points on the scale (measured by any group with an initial predisposition Our last group problem may be the most
reference to the initial median point). will inevitably be skewed in the direc- interesting of all. Suppose a group has
When members are initially disposed tion of that predisposition. As a statisti- a great deal of information—enough to
toward risk taking, a risky shift is likely. cal matter, the arguments favoring the produce the unambiguously right out-
When they are initially disposed toward initial position will be more numerous come if that information is elicited and
caution, a cautious shift is likely. A than those pointing in another direc- properly aggregated. Even so, the group
finding of special importance for busi- tion. Individuals will have thought or will not perform well if its members
ness is that group polarization occurs heard of some but not all the arguments emphasize broadly shared information
for matters of fact as well as issues of that emerge from group deliberation. while neglecting information that is
value. Suppose people are asked how Thus deliberation will naturally lead held by one or a few. Countless studies
likely it is, on a scale of zero to eight, people toward a more extreme point in demonstrate that this regrettable result
that a product will sell a certain number line with what they initially believed. is highly likely.
of units in Europe in the next year. If The second reason involves repu- “Hidden profiles” is the technical
the pre-deliberation median is five, the tation again. As we have seen, people term for accurate understandings
group judgment will tend to go up; if it’s want to be perceived favorably by other that groups could but do not achieve.
three, the group judgment will tend to group members. Sometimes their Hidden profiles are a product of the
go down. publicly stated views are a function of “common knowledge effect,” whereby
Even federal judges—experts in how they want to present themselves. information held by all group members
the law and supposedly neutral—are Once they hear what others believe, has more influence on group judgments
susceptible to group polarization. they will adjust their positions at least than information held by only a few.
Research by one of us (Sunstein, along slightly in the direction of the domi- The most obvious explanation of the

82 HBR Special Issue

Fall 2023
 Leaders can refuse to take a firm position at the
outset, thus making space for more information to emerge.

effect is that common knowledge is on discussions and conclusions. The the outset and in that way make space
more likely to be communicated to the executives gave disproportionately little for more information to emerge. Many
group. But faulty informational signals weight to valuable information held studies have found that members of
play a big role as well. by one person or a few, and as a result low-status groups—including less-
Consider a study by Ross Hightower made bad decisions. educated people, African-Americans,
and Lutfus Sayeed on how groups make The study also found that some and sometimes women—have less
personnel decisions. Résumés for three group members are “cognitively cen- influence within deliberating groups
candidates for the position of market- tral,” in that their knowledge is held (and may self-silence). Leaders who
ing manager were placed before three by many other group members, while model an open mind and ask for candid
group members. The experimenters other group members are “cognitively opinions can reduce this problem.
rigged the résumés so that one appli- peripheral,” in that their information is “Prime” critical thinking. We have
cant was clearly the best for the job. But uniquely held. To function well, groups seen that when people silence them-
each test subject was given a packet of need to take advantage of cognitively selves in deliberating groups, it is often
information containing only a subset of peripheral people. But in most groups, out of a sense that they will be punished
attributes from the résumés. cognitively central people have a dis- for disclosing information that runs
Almost none of the deliberating proportionate influence on discussion. counter to the group’s inclination. But
groups made what would have been, A simple explanation for this is that social norms are not set in stone. Social
if all the information were considered, group members prefer to hear informa- scientists have done a lot of work on the
conspicuously the right choice. The tion that is commonly held—and prefer importance of “priming”—that is, trig-
winning candidates tended to be those to hear people who have such infor- gering some thought or association in
about whom all three test subjects were mation. Cognitively central people are such a way as to affect people’s choices
given positive information. Negative thus accorded high levels of credibility, and behavior. In experiments on group
information about the winner and whereas cognitively peripheral people decision-making, engaging partici-
positive information about the losers are accorded low levels. pants in a prior task that involves either
(revealed to only one or two group “getting along” or “critical thinking” has
members) did not reach the full group. been shown to have a big impact. When
While many hidden-profile experi- Making Groups Wiser people are given a “getting along” task,
ments involve volunteers from college A central goal in group decision-making they shut up. When given a “critical
courses, similar results have been should be to ensure that groups aggre- thinking” task, they are far more likely
found among real-world managers. In a gate the information their members to disclose what they know. So if the
study of hiring by high-level executives actually have and don’t let faulty leader of a group encourages informa-
conducted by Susanne Abele, Garold informational signals and reputational tion disclosure from the beginning,
Stasser, and Sandra I. Vaughan-Parsons, pressures get in the way. Here are six even if it goes against the grain, mem-
the experimenters did not control the ways to achieve that goal, starting with bers will probably do less self-silencing.
information the executives had about the simplest: Reward group success. People
the various candidates; instead, the Silence the leader. Leaders often often keep silent because they receive
executives did their own information promote self-censorship by expressing only a fraction of the benefits of disclo-
searches. As a result, some information their own views early, thus discouraging sure. Careful experiments have shown
was known to all, some was shared but disagreement. Leaders and high-status that incentives can be restructured to
not by all, and some was held by only members can do groups a big service by reward group success—and hence to en-
one person. indicating a willingness and a desire to courage the disclosure of information.
The finding? Common information hear uniquely held information. They Cascades are far less likely when each
had a disproportionately large impact can also refuse to take a firm position at individual knows that he has nothing to

HBR Special Issue

Fall 2023 83
 BUILD A BET TER PROCESS
MAKING DUMB GROUPS SMARTER

gain from a correct individual decision with this approach: Authentic dissent but for all those affected by them. The
and everything to gain from a correct and a formal requirement of devil’s good news is that decades of empirical
group decision. The general lesson is advocacy are different; the latter does work, alongside recent innovations,
that identification with the group’s suc- far less to improve group performance, offer some practical safeguards and
cess is more likely to ensure that people because members are aware that it’s correctives that can make groups a
will say what they know, regardless of artificial—a kind of exercise or game. lot wiser.
whether it fits “the party line.” (This, Establish contrarian teams. HBR Reprint R1412F
by the way, is one reason that predic- Another method, related to appoint-
tion markets work and deserve careful ing a devil’s advocate but shown to be Cass R. Sunstein is the Robert Walmsley
attention.) more effective, is “red teaming.” Red University Professor at Harvard Law School.
Assign roles. To understand one teams come in two basic forms: those Reid Hastie is the Ralph and Dorothy Keller
especially promising strategy, imagine a that try to defeat the primary team in Distinguished Service Professor of Behav-
deliberating group consisting of people a simulated mission, and those that ioral Science at the University of Chicago
Booth School of Business. They are the au-
with specific roles that are known and construct the strongest possible case
thors of Wiser: Getting Beyond Groupthink
appreciated by all members. One person against a proposal or a plan. Red teams to Make Groups Smarter (Harvard Business
might have medical expertise; another are an excellent idea in many contexts, Review Press, 2015), from which this article
might be a lawyer; a third might know especially if they sincerely try to find is adapted.
about public relations; a fourth might mistakes and exploit vulnerabilities and
be a statistician. In such a group, sensi- are given clear incentives to do so.
ble information aggregation would be The Delphi method. This approach,
far more likely, simply because every developed at the RAND Corporation
member would know that each of the during the cold war, mixes the virtues of
others had something to contribute. individual decision-making with social
Indeed, experiments have found that learning. Individuals offer first-round
the bias in favor of shared informa- estimates (or votes) in complete ano-
tion is reduced when test subjects are nymity. Then a cycle of re-estimations
openly assigned specific roles. If a group (or repeated voting) occurs, with a re-
wants to obtain the information that quirement that second-round estimates
its members hold, they should be told have to fall within the middle quar-
before deliberations begin that each has tiles (25%–75%) of the first round. This
a different and relevant role—or at least process is repeated—often interspersed
distinctive information to contribute. with group discussion—until the par-
Appoint a devil’s advocate. If ticipants converge on an estimate. A
hidden profiles and self-silencing are simple (and more easily administered)
sources of group failure, a tempting alternative is a system in which ultimate
approach is to ask some group members judgments or votes are given anony-
to act as devil’s advocates, urging a mously but only after deliberation. An-
position that is contrary to the group’s onymity insulates group members from
inclination. Those who assume that role reputational pressures and thus reduces
can avoid the social pressure that comes the problem of self-silencing.
from rejecting the group’s dominant po- Group failures often have disastrous
sition, because they have been charged consequences—not merely for busi-
with doing precisely that. But be careful nesses, nonprofits, and governments,

84 HBR Special Issue

Fall 2023
Copyright © Harvard Business Publishing. All Rights Reserved. This content is intended for
individual research use only, subject to the following:

Unless permission is expressly granted in a separate license, this content may NOT be used
for classroom or teaching use, which includes teaching materials, electronic reserves, course
packs or persistent linking from syllabi. Please consult your institution's librarian about the
nature of relevant licenses held by your institution and the restrictions that may or may not
apply.

Unless permission is expressly granted in a separate license, this content may NOT be used in
corporate training and/or as corporate learning materials. For corporate users, please consult
the specific terms of your company's license(s) for complete information and restrictions.

For more information and teaching resources from Harvard Business Publishing including
Harvard Business School Cases, eLearning products, and business simulations please visit
hbsp.harvard.edu.