Groupthink: When a Committee Stops Challenging Itself

Groupthink is a committee losing the ability to challenge itself. Groupthink is not people agreeing. Each person who has already spoken raises the cost for the next one who wants to object, so disagreement stops being available. The failure sits in the process rather than in the people, so the correction sits there too, in the order they speak.

Start with the fact that makes groupthinkA committee losing the ability to challenge itself. so hard to catch. Groupthink rests on a fact about rooms, not a fact about characters. The cost of stating a view against the direction a room is taking rises with every person who has already committed to that direction, and it rises for everybody in the room equally, including the person who would have objected instantly over a cup of tea afterwards. Nobody in a room showing groupthink is behaving badly. Explanations built on character are useless for exactly that reason. Twelve careful, honest, well prepared people produce it just as reliably as twelve careless ones, because the thing doing the work is the order they speak in and not the quality of the people speaking.

What is groupthink, as a property of a process rather than of people?

A committee has three stages, and only one of them can be redesigned. First there is what each person privately thinks and walks into the room already holding. Then there is the process: who speaks, in what order, and whether the speaking is done aloud or on paper. Last there is the record: the stated views, the decision, and whether any objection appears in writing. Groupthink is a property of the middle stage, and the middle stage is the only one anybody controls. No rule can legislate what a colleague privately believes, and no rule can make a set of minutes report a view that was never voiced. A committee can still decide who speaks second.

Here is the everyday version. A household of five sits down to choose a school. If the eldest earner states a preference first, complete with reasons, the other four are no longer choosing a school; they are deciding whether to contradict the person who just spoke. The two decisions carry different costs, and the second one has an answer that has nothing to do with schools. Nobody at that table is weak. The table is arranged badly.

Why does the cost of objecting rise with each person who speaks?

The name comes from Irving Janis, in Victims of Groupthink, 1972, where it describes a group whose drive to stay agreeable overwhelms its ability to look at alternatives honestly. The pressure underneath it had already been measured: Solomon Asch, in Opinions and Social Pressure, Scientific American, 1955, showed people abandoning the plain evidence of their own eyes about the length of a line once enough others in the room had confidently said something else. Nothing in either account requires anybody to be timid.

Three things happen every time somebody speaks. First, a proposition acquires a person. Disagreeing with an idea is cheap. Disagreeing with an idea that Devika Rao stated forty seconds ago carries a message about Devika Rao, so it is a different act. Second, the size of the claim an objector has to make grows. Contradicting the first speaker means saying "I read it differently". Contradicting the eighth means saying "all eight of you have missed something". The eighth objection is a much larger and much more embarrassing thing to be wrong about. Third, silence gets counted as consent. Everybody who has stayed quiet is added to the apparent weight of the room, so the room looks more settled than it is, and each new person is deciding against a total that includes people who have said nothing at all.

The cost of objecting rises with the number of people who have already committed, and that rise is the entire mechanism. Notice what is not in that sentence. There is no personality, no rank, no culture and no bad actor. Take the same twelve people, ask them the same question in twelve separate rooms, and the pressure disappears completely. Whatever generated the pressure was never inside any of them.

What does the arithmetic of falling dissent look like?

A number makes the shape of the effect plain, and the shape of the number is the lesson. Take an invented committee of twelve people, all of whom privately hold the same reservation about the proposal in front of them. On the illustrative scale used here, somebody speaking first voices that reservation 70.0 per cent of the time. Each person who has already spoken takes 5.0 points off. So with three ahead it is 55.0 per cent, with five ahead 45.0, with eight ahead 30.0, and the twelfth person, with eleven ahead, voices it 15.0 per cent of the time. The private reservation is stipulated identical at every seat. Only the seat changes.

There is no threshold in that column, and the absence of a threshold is the finding. A room does not sit safely at full challenge for a while and then tip over at speaker seven. The share loses 5.0 points at every single seat, from the second person onward, in a straight line. By the eighth seat the share has fallen to 35.0 per cent, exactly half of what the first speaker had, and no single moment along the way can be blamed for it. The absence of a threshold is precisely why the seating order is worth arguing about. A tipping point could be watched for; with no tipping point, the only defence is arithmetic applied in advance.

What does a committee lose that the same people asked separately have?

Independence between the views. Independence is the whole of it, and the word needs care: independence is not honesty, and it is not courage or preparation either. A set of twelve people asked the same question in twelve separate rooms produces twelve readings that were each arrived at without reference to the other eleven. The same twelve around one table produce one reading, stated first, and eleven responses to it. The room does not collect twelve views; it collects one view and eleven adjustments to it, and the adjustments are not evidence in the way the views were.

Now count the cost with the same illustrative scale. If all twelve were somehow in the position of the first speaker, the expected number who voice the reservation is twelve multiplied by 0.700, or 8.4 of the twelve. Run them in a speaking order instead, and the expected number is the sum of 0.700, 0.650, 0.600 and so on down to 0.150, and that sum comes to 5.1 of the twelve. The room has lost 3.3 voiced reservations out of 8.4, being 39.3 per cent of the challenge that was sitting in the room when everybody walked in. The average share across the twelve seats is 42.5 per cent against 70.0 per cent for a first speaker.

The falling share is why a room's claim to have discussed it thoroughly, with everybody given a chance to speak, is such a weak defence. Everybody did have a chance to speak. The chances were not equal, and the inequality was set by the running order before anybody opened their mouth.

Which visible signs show groupthink, and which two of them mislead?

Six things are visible from outside a room, and all six are worth learning before any of them is trusted. The room converges quickly. Nobody objects on the record. The framing used by whoever spoke first comes back out of other mouths, in the same words. Alternatives get named but never examined, so a list of options appears in the minutes with only one of them costed. The real objection turns up afterwards, in a corridor, to one person. And the room starts checking how the decision will look rather than whether it is right.

Two of those six are actively misleading, and they are the two that are easiest to observe. The first is speed. A committee that converges in ten minutes has either converged because the answer was genuinely obvious, or because the second speaker made the third one expensive. Both readings fit the same ten minutes exactly, and nothing in the record separates them. One of the two is flattering to everybody present, and flattery is not a neutral fact about which reading gets chosen. The second is the absence of a recorded objection. A set of minutes showing no dissentA view stated against the direction the room is taking. was either produced by a room that held none, or by a room that made voicing it too expensive. The document is produced by the very process under suspicion, so the document cannot say which.

The same problem appears in the invented Palash decision log, a record of 240 decisions taken by 60 investors over eight quarters. A written reason appears on 84 of the 240, being 35.0 per cent. The remaining 156 are not decisions taken for no reason. The 156 are decisions whose reasons were never written down, and no amount of staring at the log will separate the two. A room's minutes have exactly this property, and it is worth naming plainly rather than treating the record as though it were a measurement.

Which arrangements restore challenge, and why do they work?

Everything that works acts on the speaking orderThe sequence in which people give their views, which changes what gets said. or on the form the speaking takes. Nothing that works acts on the people. Five arrangements are worth knowing. Have the most junior member speak first, so the expensive seat is occupied by the person whose view would otherwise have been cheapest to skip. Have whoever chairs state a position only at the end. A chair speaking first commits the room in a single sentence. Rotate the order between meetings, so no one person is permanently in the twelfth seat. Ask for every view in writing before anybody speaks. And ask each person not for a verdict but for what would change their mind. The request turns stating a doubt into answering a question rather than into an act of opposition.

Not one of those five asks anybody to be braver, and that is the sign that they are likely to work. An arrangement that depends on somebody being willing to pay the cost of the twelfth seat has not removed the cost; it has just found somebody to pay it, and it stops working the day that person is unwell. An arrangement that removes the cost keeps working whoever is in the room. The same reasoning sits behind the written checklist adopted by 20 of the 60 investors in the invented Palash log on 4 November: across the following four quarters those 20 recorded a written reason on 34 of 41 decisions, being 82.9 per cent, against 19 of 63, being 30.2 per cent, for the other 40. The checklist did not make anybody more thoughtful. The checklist changed what the act of deciding required.

Why is written before spoken the strongest single change?

Because writing every view down first does not reduce the ordering. Writing first deletes the ordering. Every other arrangement moves people around inside a sequence that still exists, so somebody is still twelfth and somebody still pays the highest cost. Written before spokenRecording a view before anybody speaks, so it cannot be revised by what the room has already heard. means there is no number of people who have already spoken to count. At the moment each view is recorded, that number is zero for everybody. On the illustrative scale used throughout, the difference is between an expected 5.1 voiced reservations out of twelve and an expected 8.4, and the twelve people are the same twelve.

The room still gets to argue, and that is the part people miss when they hear the words written down. Nothing about writing first shortens the discussion or makes it politer. Writing first changes what the discussion starts from: twelve readings arrived at independently, rather than one reading and eleven responses. The argument that follows is usually longer and certainly more uncomfortable. Disagreements that would have quietly failed to appear are now sitting on the table in twelve people's own handwriting, and somebody has to deal with them.

Where is a committee converging quickly the right outcome?

Most of the time. Rooms are not bad at deciding, and an account of how they fail reads far too easily as though they were. Most questions that reach a committee genuinely are easy, and quick convergenceThe room arriving at one position, whether quickly or slowly. on an easy question is a committee doing exactly what it exists to do. A room that spends fifty minutes manufacturing disagreement about something obvious has wasted its one scarce resource, the attention of twelve people at the same time, and it will have none of that attention left for the hard question later in the agenda.

The check is not how long it took; it is whether the question was asked before it was answered. There is one practical form of that check, and it fits in a sentence: did anybody state what would have changed their answer? If somebody named a condition, and the room looked at whether the condition held, then ten minutes was ten minutes well spent. If nobody named one, then the room has agreed on an answer to a question it never put, and the ten minutes settles nothing either way. The test is available in real time and costs one question from whoever is chairing.

How does this get used, by a committee and by somebody deciding alone?

The professional reader comes first. Devika Rao, the adviser at the invented Palash Advisory Services Private Limited, sits on a small committee that reviews the four positions in the invented holding: the Vindhya index scheme, the Nilgiri mid-cap scheme, Suvarna Chemicals Limited and Kesari Logistics Limited. On 12 October the decision taken was to sell Suvarna Chemicals Limited whole at Rs 4,60,000/-, booking Rs 60,000/-, and to keep Kesari Logistics Limited, then down Rs 1,05,000/-. Whatever the merits of that pair of decisions, what matters is what the room has to be able to show afterwards: not that everybody agreed, but that somebody said out loud what would have made the answer different. If the first person to speak framed it as taking a profit, the second person is no longer being asked about Kesari Logistics Limited at all.

The practical use is a single line added to the record: what would have changed this answer, and who said it. That line is cheap, it is checkable months later by somebody who was not there, and it fails loudly when nobody can fill it in. The pressure behind that line is not the same in every quarter. In the invented cohort, activity ran at 3.1 times its eight quarter median in the quarter when the illustrative Palash 100 index peaked at 131.0, and at 0.4 times in the quarter when the index sat at its low of 104.0. A room meets its hardest version of this problem when everything is going up and objecting sounds like pessimism.

Now the reader with no committee at all. There is no seating order, but there is something that behaves like one: the first person told. Saying it to a colleague or to somebody at home before anything has been written down makes the teller the first speaker and the listener the second, with all the arithmetic in this guide running against the listener. The defence that works alone is the defence that works in a room: writing the reason and what would change it before any of it is said aloud. Writing it down costs two minutes, and it is the only part of the defence that can be done entirely alone.

Which questions does groupthink not answer?

Three, and each is worth separating from what groupthink does explain. Groupthink says nothing about why the person who speaks first holds the view they hold; whatever produced that view was already complete before the meeting started, and the mechanism starts with what happens to the eleven views afterwards. Groupthink says nothing about a decision taken by one person with nobody watching and nobody to tell. Such a decision has its own pressures and none of these. And groupthink is not a claim that a committee is worse than an individual. A room contains twelve readings and usually hears only one of them, so the argument is for changing the room rather than for abolishing it.

A committee picked out after the fact because its decision went wrong cannot be examined honestly. The outcome selected the case, so every feature of that room reads afterwards as a cause: the speed, the silence, the seating order. A committee whose dissent shares are stipulated in advance carries no such selection, and it can therefore be examined.

Devika Rao, Palash Advisory Services Private Limited, the Palash decision log, the Palash 100 index, the Vindhya index scheme, the Nilgiri mid-cap scheme, Suvarna Chemicals Limited, Kesari Logistics Limited and the twelve-person committee with its dissent shares are invented.
Educational material. Not advice on any investment, tax, budget or market position.