The shirt had a large picture of Barry Manilow on the front, and that was entirely the point.

In the opening experiment of The Spotlight Effect in Social Judgment, published in the Journal of Personality and Social Psychology in February 2000, Thomas Gilovich, Victoria Husted Medvec and Kenneth Savitsky recruited 109 Cornell undergraduates. Fifteen were handed the Manilow shirt, told to put it on, and sent into a room where four or five other students sat filling out questionnaires. Within moments an experimenter pulled them back into the hallway and asked one question: how many of the people in that room could say who was on your shirt?

The average estimate was 46 per cent. Observer accuracy was 23 per cent. Comparing each wearer’s guess against the observers actually present in that session gave a mean overestimate of 23 percentage points, with a 95 per cent confidence interval running from 9 to 38.

A separate group of thirty students watched a staged video reconstruction of the same scene and gave much lower estimates, which suggests the wearers were not simply working from a pessimistic general theory about how observant people are. Something about being the one in the shirt did the work.

Fifteen wearers, sixty-four observers, one American campus in the 1990s. The scale of the gap belongs to this sample; a different generation, a different culture, or a milder embarrassment might return a different number.

The flattering shirt produced the bigger gap

Study 2 removed the embarrassment. Fifteen new participants chose a shirt bearing Bob Marley, Jerry Seinfeld or Martin Luther King Jr., and rated themselves pleased to be seen in it. They estimated 48 per cent. Observers managed 8 per cent.

The researchers suggest a mundane reason for that low figure: the questionnaire observers filled out in this study was more absorbing than the one used in Study 1, which likely pulled attention from the doorway. The wearers’ estimates barely moved between conditions all the same. Pride and mortification produced almost the same number.

Study 3 moved from clothing to conduct. One hundred and ninety-three students, in forty-two groups, spent twenty minutes discussing a policy question and then ranked themselves as they believed the group would rank them. On all six measures, including speech errors and comments that might have offended someone, participants placed themselves more prominently than their fellow discussants actually did.

A 1979 paper had already pointed this way

The authors are explicit that they are extending Michael Ross and Fiore Sicoly’s work on egocentric biases in availability and attribution, five experiments run with discussion groups, married couples and intercollegiate basketball teams, which found that people recall their own contributions to a joint effort more readily and claim more credit for an outcome than their collaborators grant them. Gilovich, Medvec and Savitsky measured a further step: what people assumed others had noticed, rather than what people themselves recalled.

Their proposed mechanism is anchoring and adjustment. A person starts from how loud the thing is in their own head, then adjusts down for the fact that other people are occupied elsewhere, and that adjustment falls short.

Two further studies test that mechanism directly. Forty-four Northwestern students wore the T-shirt and were then asked a simple question: before settling on their estimate, had they considered any other number? Thirty-two said yes. Of the thirty whose alternative sat clearly above or below their final answer, twenty-three had first thought of something higher.

A companion study varied one thing only: how long participants sat with the shirt before meeting anyone. Walk in straight away, and observers were expected to spot Manilow 51 per cent of the time. Wait fifteen minutes first, long enough to grow used to wearing it, and that figure fell to 37.

What the paper does not show

It does not show that nobody notices anything. In Study 1, wearers’ estimates correlated with actual observer accuracy at r = .50, and in the group discussion every one of the six correlations was substantial. People who thought they had talked a lot had in fact talked a lot. People’s sense of the ranking held up well across the board; the number that ran high, in every study, was the absolute level of attention they expected, not their place in the order.

The authors also describe the opposite case. When behaviour becomes automatic, self-focus drops and the effect can flatten or invert. Their examples are smokers and the heavily perfumed, both of whom may underestimate how noticeable they are to a room.

And the samples are small and specific: fifteen shirt wearers per study, undergraduates earning course credit, confidence intervals wide enough that the honest summary is a substantial gap of uncertain size rather than a precise ratio.

Where the finding has since been put to work

Savitsky, Nicholas Epley and Gilovich followed it in 2001 with four studies on social blunders, finding that people who imagined or experienced an embarrassing failure expected harsher judgement than observers delivered. Two years later, Savitsky and Gilovich reported that simply telling nervous speakers about the illusion of transparency, the related tendency to assume internal states leak out, reduced how visible they believed their nerves were and improved how their speeches were rated in that experiment. The illusion of transparency was itself set out in a 1998 paper by the same three authors.

Gilovich and colleagues close their discussion with a broader claim they don’t pretend to have tested: some of what people never do, the singing, the dancing, the joining in, gets declined out of a fear of judgement that the research suggests runs higher than warranted. It’s a hypothesis the paper makes plausible, nothing more.

What stays with us is the fifteen minutes. Same shirt, same face on it, same room of strangers waiting. Sitting with it a while was enough to knock fourteen points off what the wearer expected the room to see.