My grand-aunt kept a jar of coins by the front door for the boy who delivered the papers, and she would not walk past the bakery at the end of her street.
The bakery was new. So was the family running it, who had come from a country she could not pronounce and who kept, in her telling, peculiar hours. She took the long way round for eleven years. Four extra minutes, on a trip she made most days. Nobody ever got a straight answer out of her about why, because there wasn’t one. She was not a hard woman. She drove neighbours to hospital appointments and fed every stray cat in the lane. Her feet simply would not go down that stretch of footpath.
I thought about her the first time I read the brain imaging work on this, because it offered something more useful than a verdict.
What the scans picked up
In 2011 a team at University College London ran structural MRI on a group of young adults and matched the images against how those people described their own politics. Writing in Current Biology, Ryota Kanai and his colleagues reported that greater conservatism went with more grey matter in the right amygdala, the almond-shaped structure that handles threat and emotional salience. Greater liberalism tracked instead with volume in the anterior cingulate cortex, a region tied to error monitoring and holding contradictory information in mind without panicking.
One paper is one paper. That one had 90 participants and a 28-person follow-up, which is a decent effort for 2011 and thin by current standards.
So the interesting question is what happened when someone tried again with a much larger sample. In 2024, researchers using the Amsterdam Open MRI Collection, 928 brains, ran a preregistered replication in iScience. They recovered a small positive relationship between amygdala volume and conservatism. The anterior cingulate half of the story did not survive. That is the shape of a real but modest effect, solid enough to report and useless for diagnosing any individual.
A bigger amygdala is not a more frightened one
Here is where the popular version goes wrong. Amygdala volume also predicts something that sounds nothing like fear at all: Kevin Bickart and colleagues found in Nature Neuroscience that people with larger amygdalae maintained larger and more complex social networks. A 2024 attempt to replicate it, which scanned 128 brains at 7 tesla and was published in Cognitive, Affective, and Behavioral Neuroscience, found only weak and inconclusive support for the real-world version of that link. Bigger still leans towards busier, just with less confidence than the original study claimed. The structure is less a fear centre than a significance detector, flagging whatever in the environment needs attention right now, and other people are the loudest thing most of us encounter.
My grand-aunt had a phone book of a social life. She knew the business of everyone on that street, which is possibly why one unknown household read to her as a gap in the map rather than a neighbour.
The firing part
Volume is anatomy. Response is a different measurement, and the evidence there goes back a quarter of a century. Using fMRI at Massachusetts General Hospital, Allen Hart and co-authors recorded stronger amygdala signal to out-group faces than in-group faces, but only during later presentations. Habituation explains the gap: signal to same-group faces faded as the images repeated, while signal to the other group held up.
That distinction matters for anyone tempted to read a scan as a character reference.
Where the threat story ran into trouble
A famous 2008 study seemed to nail the whole thing down. Douglas Oxley and colleagues, publishing in Science, measured skin conductance and blink reflexes in 46 politically committed adults. Those who reacted most sharply to sudden noises and threatening images were likelier to back defence spending and capital punishment. The calmer responders leaned towards foreign aid and liberal immigration policy.
Then other laboratories tried it. In 2020, Bert Bakker and three co-authors published a direct preregistered replication plus two conceptual ones, 635 participants in total across the United States and the Netherlands, and reported in Nature Human Behaviour that the effect came up empty. Same images, four times the sample, no result.
So the honest summary is a small structural correlation that has replicated, a response difference to unfamiliar faces that looks like uneven habituation, and a physiological threat-sensitivity story that has not held up. Together they give the discomfort a thin biological thread, and that’s about the size of it.
What familiarity does to the signal
Elizabeth Phelps, at New York University, showed that amygdala response to other-race faces tracked people’s implicit bias scores, with one qualification that I keep returning to. When participants were given faces they already knew, the relationship disappeared. A known face comes with a history attached, and the machinery appears to treat it accordingly.
Which lines up with the largest body of evidence anyone has assembled on this: Thomas Pettigrew and Linda Tropp’s meta-analysis of 515 studies concluded that contact between groups generally reduces prejudice, and that the effect travels beyond the individuals you actually meet.
My grand-aunt broke her wrist in her late seventies. The bakery family’s eldest daughter, who happened to be a nurse, drove her to the clinic and sat with her for three hours. After that my grand-aunt used the short way, told anyone who asked that the sourdough was the best on the street, and never once mentioned the eleven years.
Her amygdala did not shrink. It just got some new information about who counted as a stranger.
