Group projects

Why does a group of capable students so often hand in work worse than its best member alone would?

Note how strong the complaint actually is. Nobody is surprised that four students produce less than four times one student's work — some duplication is expected. The complaint is sharper and stranger than that: the finished report is worse than the best member would have written alone. Adding three capable people made the artefact worse. What could possibly have that shape, given that nothing was subtracted?

The staffroom answer is that some of them were lazy, or that they clashed. Both are testable, and both turn out to be poor explanations.

Take laziness first. Ingham and colleagues in 1974 blindfolded people on a rope and told them either that they were pulling alone or that others were pulling with them. Nobody else ever pulled. Yet participants who believed they were part of a group pulled measurably less hard — on the order of a tenth less — than the same people who believed they were alone. Same person, same rope, same muscles. Whatever was lost was not a trait; it was produced by the belief that effort was shared.

That study is worth pausing on because it settles something the older evidence could not. Ringelmann, around 1913, had found that eight men on a rope produced roughly half the sum of their individual pulls — but a rope mixes two very different losses. Eight people pull slightly out of phase and in slightly wrong directions, which is coordination loss and has nothing to do with willingness. Ingham's blindfolded pseudo-groups had no one to be out of phase with, so what remained had to be motivation.

Now ask why motivation should fall, because "people slack in groups" is a restatement, not a mechanism. Look at what the shared grade does to the arithmetic each member faces. My effort raises a mark that everyone receives; the tiredness is mine alone. So when I decide how much to do, I weigh my whole cost against my own share of the benefit and ignore the benefit flowing to the other three. Every member reasons the same way, so every member under-supplies — and the size of the ignored benefit grows with the group. That predicts the shortfall should deepen as groups get larger, and it does: in Latané, Williams and Harkins's 1979 shouting studies, individual output fell to around seven-tenths in pairs and around two-fifths in sixes.

But under-supplied effort still does not explain falling below the best member. For that, ask how the pieces combine. Steiner's useful distinction is between tasks where the group's output is the best member's answer, tasks where it is the sum, and tasks where it is the weakest link. A group report is not one of these — it is divided into parts, and it is read as a whole. Inside a section, the best member's work stands on its own merit. Across the seams, the reader's judgement of coherence, argument and consistency is set by the worst joint. So the artefact behaves like a best-member task locally and a weakest-link task globally, and the global reading is the one that gets a mark.

Two further losses arrive in the meeting itself. Stasser and Titus showed that discussion gravitates toward information everybody already holds and neglects what one member uniquely knows — so the best member's distinctive contribution is exactly the sort least likely to be adopted. And Diehl and Stroebe demonstrated in 1987 that groups generating ideas together produce fewer than the same number of people working separately, mostly because only one person can speak at a time and the others lose their thought while waiting.

Let me try to break my own account. If it were right, groups should sometimes beat their best member — specifically on problems where a correct answer can be demonstrated to the others once someone states it. Then the group only needs one person to see it and the rest to recognise it, and the weakest-link seam never forms. That is what happens: on such problems groups reliably outperform their best individual. The account survives because it predicts the exception as well as the rule.

A refinement that connects the two halves: motivation loss and structural loss are independent, and they need different fixes. Making contributions individually visible — separately assessed sections, tracked drafts, a viva — attacks the first and does nothing for the second. Assigning an integrator, or grading the seams explicitly, attacks the second and does nothing for the first.

Worth flagging as contested: how much of the rope-and-shout deficit is motivation and how much is coordination is still argued, and the classic Ringelmann figure is the shakiest thing here — it comes from a thin nineteenth-century demonstration mostly known through later retellings. The direction is not in doubt; the magnitude should be held loosely.

Note also what this is not. The familiar observation that adding people to a late project makes it later concerns communication channels multiplying with headcount. That is a different quantity: it explains why the group is slow, not why its output is worse than one member's.

• Why groups outperform their best member on problems with demonstrable answers, and what makes an answer demonstrable.
• Whether individually assessed components recover the loss, or merely relocate it into the seams.

Every term the collection defines is gathered in the glossary.