Confidence appears to be contagious. A Birkbeck study found that interacting with a confident partner changed people's own sense of certainty — and they stayed more confident even after the social interaction ended and they were making decisions alone again.

Every workplace has its own acceptable volume of certainty. In one team, saying “I think this is right” sounds appropriately careful. In another, the same sentence is heard as weakness, while the person who speaks without qualification becomes the centre of gravity.

Some of that is performance. We adjust how certain we sound because we want to persuade, avoid embarrassment or fit the room. A new study asks a more unsettling question: after matching someone else’s confidence in public, do we carry a changed sense of certainty back into our private decisions?

In Current Biology, Einar Andreassen, Chris Frith and Birkbeck’s Daniel Yon report that participants did. People paired with a confident decision partner became more confident themselves. Those paired with an uncertain partner became less confident. The difference remained when the partner vanished and participants went back to deciding alone.

This is one study, not settled consensus.

It measured confidence in a tightly controlled visual task, not general self-belief, charisma or courage.

The “partner” was a carefully controlled illusion

Across two online experiments, 204 UK adults aged 18 to 50 watched brief clouds of moving dots and decided whether they were travelling left or right. They used a sliding scale to report both their choice and how certain they felt. High confidence won more points when the answer was correct and lost more when it was wrong, so careless boasting carried a cost.

Each experiment had three consecutive phases of 90 trials: alone, together and alone again. During the middle phase, participants believed another person was judging the same display. They saw that partner’s choice and confidence, and the more confident answer became the pair’s joint decision.

There was no other person. The partner was an algorithm programmed to sound generally confident or uncertain. Both versions were given the same overall accuracy as the real participant. What changed was the confidence signal, not the partner’s competence.

The first experiment included 80 people. The second was a preregistered replication with 124, an important detail because the main pattern appeared twice rather than emerging from one exploratory sample.

Certainty moved towards the partner, then lingered

While making joint decisions, participants shifted towards the algorithm’s style. Those shown high confidence raised their own ratings. Those shown low confidence reduced them. This builds on earlier research showing confidence matching during group decisions.

The more interesting test came immediately afterwards. Once the screen stopped showing a partner and people returned to the solo task, their confidence did not simply reset. The groups remained separated: prior exposure to the confident partner was associated with higher reported certainty, and prior exposure to the uncertain partner with lower certainty.

“Contagious” is a fair shorthand, provided its timescale stays visible. The researchers showed persistence into the next block of the same experiment. They did not test whether the effect lasted through the afternoon, let alone for weeks, or whether confidence about moving dots travelled into other areas of life.

The study changed confidence, not competence

More confidence is not automatically better judgement.

The confident and uncertain algorithms were equally accurate. The researchers also examined metacognitive sensitivity, which is the ability to assign more confidence to correct choices than to errors. Partner type did not measurably improve that skill. People still responded to easier and harder trials, but their overall level of certainty had been nudged.

This distinction is easy to lose in everyday talk about confidence. Confidence can help a person act, speak or decide, but calibration determines whether that feeling tracks the available evidence. Someone can be consistently sure and poorly calibrated. Someone else can sound cautious while discriminating accurately between what they know and what they do not.

The experiment deliberately pulled these qualities apart. Participants absorbed some of the partner’s confidence even though that confidence contained no additional information about accuracy.

A cautious reading for workplace decisions

This was not an office study. It cannot tell us that one forceful manager will permanently change a team’s psychology, and it offers no basis for choosing colleagues as a confidence treatment.

It does, however, make workplace confidence cultures harder to dismiss as mere theatre. Silicon Canals has written before about rooms that mistake speaking most for thinking best. If expressed certainty also changes the baseline from which other people judge themselves, the loudest norm in a room may influence more than who wins one discussion.

An uncertain partner pulled confidence down as reliably as a confident one pushed it up. That is not necessarily bad. Caution can be appropriate when evidence is thin, just as confidence can be appropriate when it is strong. The organisational problem begins when a group’s preferred style of expression becomes detached from the quality of its decisions.

What the computational model can and cannot explain

Andreassen and colleagues built a model in which people maintain an expectation about how confident they ought to feel. That expectation is updated using discrepancies between anticipated confidence and the confidence reported by both oneself and the partner. The model reproduced the observed rise or fall during the joint phase and its carry-over into the solo phase.

It is an explanation of the behavioural pattern, not a recording of a biological process. The experiments did not use brain imaging, and another model might also account for the data.

The effect also did not vary according to whether participants later said they believed the partner was a real person, doubted it or were unsure. That result hints that repeated confidence signals may be enough to shift the internal reference point, without a rich relationship between two people. But it could equally reflect the unusual simplicity of an online task in which a number appears 90 times.

The unanswered question is duration

The Birkbeck account of the research is explicit that further work is needed to learn how long socially induced changes last. Real interactions also contain explanations, status differences, trust and feedback, none of which was captured by a programmed partner with a confidence slider.

The result is still useful precisely because the experiment stripped so much away. The partner did not know more. It was not more accurate. It simply expressed a different level of certainty, repeatedly, and people adjusted towards it. For at least the decisions that followed, some of that borrowed confidence remained.