Why unanimous agreement lowers confidence in a verdict

A panel of twenty experts agrees unanimously on a difficult question. Intuitively this is the strongest possible endorsement. Statistically it is suspicious: if the question is genuinely difficult, some dissent is expected, and its absence implies something other than twenty independent correct judgements.

What everyone sees

Unanimity is read as convergence on the truth: if everybody agrees, the answer must be right. The more people who agree, the stronger the evidence. This reading treats each expert as an independent source, and assumes that agreement scales linearly.

What is actually happening

In Bayesian terms, perfect agreement among many independent sources on a low-probability event is less likely than imperfect agreement, because each additional independent agreement multiplies the prior against a rare outcome. Condorcet’s jury theorem holds only when voters are independent and each is more likely right than wrong; if they are correlated — sharing the same training, the same data or the same social pressure — unanimity reveals the correlation rather than the truth. Sunstein documented how group deliberation amplifies shared biases, producing consensus that reflects the group’s composition more than the question’s answer.

Why it stays hidden

The paradox hides because unanimity feels like the strongest signal, and questioning it sounds perverse. A dissenting voice is treated as noise, and its absence as clarity. The logic that makes perfect agreement suspicious — that it implies correlated error rather than independent correctness — runs against the intuition that more agreement is always more evidence, and the intuition is what is reported in everyday reasoning.