The Salmon of Doubt*

I don’t think I am particularly good at reading social situations, but I get by. But I would guess that I am better at reading social situations than a dead salmon. A few years ago, Bennett and coworkers used the latest hi-tech method to study brain activity, functional magnetic resonance imaging (fMRI), to study the brain of dead salmon. While they were using the brain imaging technique on the brain of the deceased salmon, they showed it pictures of groups of people in different social situations.

They found correlations between detected activity in parts of the salmon’s brain and whether the picture showed an inclusive or exclusive social group of people. You may find this surprising. How can a dead salmon judge a social situation? We have all at some point probably misread a social situation, it would be disappointing if we were beaten by a dead fish.

So how come the dead fish’s performance looks so impressive? The answer lies in the fact that fMRI scans collect a lot of data when they image a brain (dead or alive), and there is substantial noise in the signal. If you collect a small amount of data which is just noise, it is very unlikely that the measured noise will be correlated with, in this case the social situation shown to the dead salmon. But if you collect a huge amount of data and look at it thoroughly, it is almost inevitable that some part of the noise will show a correlation. It would be very unlikely for none of the noise to be a bit higher than usual when the salmon was showed one social situation, and a bit lower when the salmon was shown the other.

Bennett et al knew this but were worried about other people using fMRI and publishing work showing correlations in their voluminous data, without being careful enough to look at the probability that this correlation arose just by chance. So they choose to highlight this problem in a clear way, using the dead salmon’s dead brain. It is straightforward here to realise that the correlation must be just by chance.

fMRI is becoming more widespread, and it is not the only technique to produce huge amounts of data. So Bennett et al.‘s work is a timely reminder that acquiring a huge amount brings risks and benefits. It also a good excuse to get lines like “It is not known if the salmon was male or female, but given the post-mortem state of the subject this was not thought to be a critical variable.” into a paper.

* The title is taken from the name of the collection of Douglas Adams’ work. See here for some entertaining quotes from that collection.