Are two heads better than one?
John Mauldin's Outside the Box

Blog Subscription Form

  • Email Notifications



This week's letter is from my favourite behavioural finance thinker, James Montier of Dresdner Kleinwort Wasserstein in London. James wrote a fascinating book two years ago called "Behavioural Finance: A User's Guide" and puts out ongoing research like the one we will enjoy today. Long time readers will recognise the name because I have discussed many of his ideas in my weekly letter "Thoughts From the Frontline" and my book "Bull's Eye Investing."

In the past we have looked at individual behavioural biases. This report by James explores why groups of otherwise intelligent people often make worse decisions as a group. You would think that the collective minds of a committee, sharing data and experiences, should perform better than the individual. But that is not always the case. A Camel, it is said, is a horse designed by a committee. Many of the examples will probably give you an "aha" moment, and in the end some strategies to overcome group behavioural biases are suggested. Let's see if James can make you think "Outside the Box."

Are two heads better than one?

Do you conduct your asset allocation via committee? Or perhaps you sit on a stock selection committee? If so perhaps you should read on. Psychologists have spent many years documenting the fact that group decisions are amongst the worst decisions ever made. Effectively endorsing the view that committees are groups of people who keep minutes but waste hours! The key reason for this appears to be that when we come together as a group we have to deal not only with our own biases, but also to cope with everyone else's biases as well.

The well-informed reader might point out that James Surowiecki has recently published a book entitled The Wisdom of Crowds1. The book purports to show that "the many are smarter than the few". The basic idea of the book is that groups outperform individuals in decision-making. Surely this is in direct contradiction of my opening paragraph?

However, Surowiecki is correct that groups outperform individuals but only under a very strict set of circumstances. Psychologists have shown that statistical groups can outperform individuals and deliberative groups. A statistical group effectively involves asking a large number of people what they think the answer is and taking the mean.

Such groups have a good track record when it comes to forecasting.

For instance, when judging the number of beans in a jar, the group average is almost always better than the vast majority of the individual members. In one such experiment, a group of 56 students was asked about a jar containing 850 beans; the group estimate was 871, a better guess than all but one of the students2.

In another experiment, a group was asked to rank ten piles of buckshot, which were only minimally different in size from each other. The group's guess was 94.5% accurate, far more so than that of almost all the group members3.

However, for a statistical group to be a useful device three conditions must be met:
  1. People must be unaffected by others' decisions (effectively their errors must be uncorrelated)
  2. The probability of being correct must be independent of the probability of everyone else being correct
  3. The participants must be unaffected by their own vote possibly being decisive.
If these conditions are broken the group's advantage is quickly lost. This is a particular cause for concern from a behavioural point of view. Critics of behavioural finance argue that if one person is overoptimistic and one is pessimistic then their mistakes cancel out. One of the foundations of behavioural finance is that people, generally, err in a similar fashion. Hence, the mistakes they make are highly unlikely to be uncorrelated.

In addition, groups can be subject to anchors just like individuals (see Global Equity Strategy, 27 August). An anecdotal story may help illustrate this point. Recently, my colleagues and I went to the Oval to watch England vs. the West Indies. At about 5pm, we decided to run a book on the total score for England's first innings. The run total was around 250 at that point. I went first and chose 350; the subsequent bets were surprisingly tightly clustered around the initial estimate. The group was at risk of anchoring. The eventual outturn was an England first innings of 470!

The other weak spot of statistical groups appears to be when the members have no idea of the answer to the question. For instance, Professor Sunstein4 asked his fellow faculty the weight, in pounds, of the fuel cell that powers the space shuttle. The actual answer is 4 million pounds. The median response was 200,000; the mean was 55,790,555 (driven by a single outlier). Both answers are wildly inaccurate.

However statistical groups are not a good description of the way in which the average asset allocation committee reaches decisions. Instead, the stereotypical asset allocation committee is formed of the regional heads of equity (i.e. Europe, US, Japan and perhaps Global), the head of fixed income, the CIO and an economist/strategist. This group then sits down and debates the issues before arriving at a conclusion. This is the prototypical set up for what psychologists call deliberative groups.

The eternal hope is that such groups will come together, exchange ideas, each bringing something different to the discussion. The aim is of course to uncover all the information the group has. If one member has an irrelevant anchor, then the hope is that the group will expose this anchor, hence beating the bias.

However, the reality of group behaviour is very different. As MacCoun5 notes "Groups generally can be expected to amplify rather than correcting individual bias" (sic). For instance, psychologists have shown that, in general, deliberation tends to reduce variance. After talking together, the members of a group will tend to reach a consensus; hence the variance of views is diminished.

Additionally, group discussion tends to lead to group members having more confidence in their decisions after the group deliberations6. However, sadly, this increased confidence is not matched by increased accuracy. People simply become more sure about the views they hold rather than enjoying an improvement in performance.

This confidence seems to be driven by the simple repetition of the view. The more you hear a view, the more you tend to have confidence that the view is correct!

Sunstein (op cit) notes "Groups have been found to amplify, rather than to attenuate, reliance on the representativeness heuristic; to reflect even larger framing effects; to show more confidence than individuals; ... In addition, groups demonstrate essentially the same level of reliance on the availability heuristic"7.

Deliberative groups also show an alarming inability to uncover information that isn't common knowledge, and instead end up centred on the knowledge that is easily available to all the group members.

Sunstein cites a wonderful example from Hightower and Sayeed8:
The purpose of the study was to see how groups might collaborate to make a personnel decision. Resumes for three candidates, applying for a marketing manager position, were placed before group members. The attributes of the candidates were rigged by the experimenters so that one applicant was clearly the best for the job described. Packets of information were given to subjects, each containing a subset of information from the resumes, so that each group member had only part of the relevant information. The groups consisted of three people, some operating face-to-face, some operating on-line. Almost none of the deliberating groups made what was conspicuously the right choice. The reason is simple: They failed to share information in a way that would permit the group to make that choice. Members tended to share positive information about the winning candidate and negative information about the losers. They suppressed negative information about the winner and positive information about the losers. Hence their statements served to "reinforce the market toward group consensus rather than add complications and fuel debate."
The general finding from a wide variety of such experiments is that unshared information is highly likely to be omitted from the discussion. Instead, members of the group will tend to concentrate on shared information leading to a hardening of view, creating an anchor for the subsequent discussions.

Why are groups so bad at uncovering unshared information? In part it is a function of the statistical fact that shared information is more likely to be discussed and repeated. Of course, this will tend to influence the individuals in the group, and hence influence the group's eventual outcome.

The other factor that helps explain why groups are so bad at sharing information concerns social pressure. Those who perceive themselves to be of a relatively low status (admittedly not a major problem for many in our industry!) are particularly likely to refrain from sharing unique information9. They are scared of looking stupid in front of their colleagues. The reputational risk is simply too high for low status members to chance sharing information if it doesn't conform to the group10.

Wittenbaum et al11 have also shown those who share information which confirms the group's views are seen as competent and more credible by their peers and by themselves! A situation Wittenbaum et al call mutual enhancement.

The chart below shows the result of one of Wittenbaum et al's experiments. Once again information over candidates for a job was the topic. After hearing from the various members of the group, people were asked to rate the other members and themselves using a 0-9 scale (strongly disagree to strongly agree) on the following two questions (i) I feel competent at determining the better job candidate, (ii) the others are competent at determining the better job candidate.

Evidence of mutual enhancement would arise if participants evaluated themselves and the others more favourably when the information was common/shared by the group (i.e they don't reveal unique information). And, when people relied on unique information, they rate themselves lower and the group would also rate them lower rather than on unique information. That was exactly what was uncovered.

The third major problem that groups encounter is the tendency to suffer cascades. We have discussed cascades in the context of markets before (see Global Equity Strategy, 13 January 2004 and 19 May 2004). However, cascades can also effect group decision processes.

A cascade is a situation whereby an individual's action is potentially independent of their private information, and totally dependent upon the observation of other's actions and/or words.

One of the key features of cascades is their tendency to exhibit idiosyncrasy. That is to say the behaviour resulting from signals of just the first few individuals drastically affects behaviour of numerous followers. Effectively, cascades are highly path dependent.

Within the context of group discussions, once an opinion has been voiced then it becomes increasingly unlikely that others will argue against it. Not only for the social reasons outlined above, but because they might assume that the person voicing the belief has more information than they do.

Cascades have proven easy to create in laboratory environments12. Players are told that one of two pots containing three balls is being used. One pot contains two red and one white ball (pot A), the other two white and one red ball (pot B). Players are selected in a random order to announce which pot is being used after having received a private signal in the form of a draw from the pot.

The results from one of their experiments are shown below. The pot used was B. However, players 3 onwards decided that the other players knew more than they did and hence ignored their private signal, despite the fact that going with the private signal would be correct on 2/3rds of occasions!

The fourth hurdle that deliberative groups must attempt to overcome is the risk of group polarization and possibly groupthink. Group polarization is the tendency for members of a group to end up in a more extreme position in line with their original beliefs after talking to one another. The increased confidence in view mentioned earlier, begins to create feedback into the extremity of view, generally creating a loop of increased confidence in more and more extreme views.

Even more extreme than group polarization is groupthink. The term was coined by Irving Janis in 1972. In his original work Janis cited the Vietnam War and the Bay of Pigs invasion as prime examples of the groupthink mentality. However modern examples are all too prevalent. The recent Senate report on the intelligence gathering by the CIA over the war in Iraq explicitly accused the CIA of displaying many of the elements of groupthink.

Groupthink is often characterised by:
  • A tendency to examine too few alternatives
  • A lack of critical assessment of each other's ideas
  • A high degree of selectivity in information gathering
  • A lack of contingency plans
  • Poor decisions are often rationalized
  • The group has an illusion of invulnerability and shared morality
  • True feelings and beliefs are suppressed
  • An illusion of unanimity is maintained
  • Mind guards (essentially information sentinels) may be appointed to protect the group from negative information

Beating the biases

Can these various problems be overcome? As with all biases the solutions are never easy to implement. However, three possible routes to reducing group biases are:

Secret ballots

The use of secret ballots obviously reduces the risk of group members coming under social pressure. So perhaps before the meeting starts members should write down their views and their preference for asset allocation, then count the votes, and debate the outcome if it is really necessary - bearing in mind the dangers inherent in this process.

Devil's advocates

Appointing a devil's advocate13 may help. However, all too often the person selected may not truly believe the role they are asked to play, and hence not really try too hard to prevent the group reaching its consensus decision. Selecting prickly disagreeable individuals with a strong contrarian view and the ability and desire to argue on almost anything would be perfect. But such individuals are hard to find, and don't fit easily into most corporate cultures.

Respect for other group members

The other factor that can help reduce the dangers of group decisions is when the group members are acknowledged to be experts in their field, and hence disparate viewpoints are easier to deal with and unshared information may be easier to uncover. However, all too often people tend to believe that they know best on almost every subject and hence tend not to display respect for the views of others.

1 A client and friend sent me a signed copy of this book. Thank you Rob.
2 Surowiecki (2004) The wisdom of crowds
3 Bruce (1935) Group judgements in the field of lifted weights and visual discrimination, Journal of Psychology
4 Sunstein (2004) Group judgements: deliberation, statistical means and information markets, Chicago Law School working paper
5 MacCoun (2002) Comparing micro and macro rationality, in Judgements, decisions and public policy (ed) Gowda and Fox (also available from
6 Heath and Gonzalez (1995) Interaction with others increases decision confidence but not decision quality, Organisation behaviour and human decision process 305
7 References include Stasson et (1988) Group consensus approaches to cognitive bias task, Japanese Psychological Research 68; Sniezek and Henry (1989) Accuracy and confidence in group judgement, Organisational behaviour and human decision processes;
8 Hightower and Sayeed (1995) The impact of computer-mediated communication systems on biased group discussion, Computers in Human Behaviour
9 Wittenbaum and Park (2001) The collective preference for shared information, Current directions in psychological science, 10 and Stasser and Titus (2003) Hidden profiles: a brief history, Psychological Inquiry, 14
10 We will explore a related issue on obedience to authority in a future weekly.
11 Wittenbaum, Hubbell and Zuckerman (1999) Mutual Enhancement: Toward an understanding of the collective preference for shared information, Journal of Personality and Social Psychology
12 For a very up to date survey see Anderson and Holt (2004) Information Cascade Experiments, forthcoming in The handbook of results in experimental economics, Smith and Plott (Ed.)
13 The term comes from the process of canonization. When the Pope considers conferring sainthood, a devil's advocate is appointed unofficially to argue against the decision.
© Dresdner Kleinwort Wasserstein Securities Limited 2004


I hope you have found these examples of group behavioural biases enlightening and help you the next time you are part of a decision making group avoid some of the pitfalls by thinking "Outside the Box."

Your ready for his next group meeting analyst,

John F. Mauldin


John Mauldin is president of Millennium Wave Advisors, LLC, a registered investment advisor. All material presented herein is believed to be reliable but we cannot attest to its accuracy. Investment recommendations may change and readers are urged to check with their investment counselors before making any investment decisions.

Opinions expressed in these reports may change without prior notice. John Mauldin and/or the staffs at Millennium Wave Advisors, LLC and InvestorsInsight Publishing, Inc. (InvestorsInsight) may or may not have investments in any funds, programs or companies cited above.


Communications from InvestorsInsight are intended solely for informational purposes. Statements made by various authors, advertisers, sponsors and other contributors do not necessarily reflect the opinions of InvestorsInsight, and should not be construed as an endorsement by InvestorsInsight, either expressed or implied. InvestorsInsight is not responsible for typographic errors or other inaccuracies in the content. We believe the information contained herein to be accurate and reliable. However, errors may occasionally occur. Therefore, all information and materials are provided "AS IS" without any warranty of any kind. Past results are not indicative of future results.

Posted 10-11-2004 4:09 AM by John Mauldin