Advancing science by showing a dead salmon pictures of people, and asking it to characterise the emotions shown

Salmon sashimiBennett, Baird, Miller and Wolford won the 2012 IgNobel prize in Neuroscience with their work with a dead (definitely dead, not just resting) salmon. There is a serious point to the work: The problem of false positives. This is a problem in statistics and is a real trap for the unwary.

They also had a bit of fun with lines in the paper 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.” An excellent line. The paper was published in the excellently named Journal of Serendipitous and Unexpected Results.

The problem the paper was bringing attention to is easy to explain because we have all come across simple examples of the sort ‘false positives’ Bennett et al were interested in. This where you are right just by chance. Sometimes with some friends I am in a pub and we have a go on one of the quiz machines in pubs, and of course for some of the questions we have no clue, and so guess. Most of the time we guess wrong but sometimes we guess right and get the correct answer just by chance.

Now, modern brain imaging can yield a huge amount of data about the activity in a brain, but our brains are very poorly understood so we can be reduced to looking for correlations in this activity with some brain function, to try and work out how our brains work. The problem comes if you measure the activity in 10,000 places in a brain at the same time, and do this while showing the brain’s owner pictures of sad people, happy people, angry people, etc.

Then if you look for correlations between brain activity and the pictures you need to be careful, if you measure 10,000 things that is like having 10,000 guesses – there is bound to be least a few that are correlated. But that could be simply be chance, it could mean nothing. You could image anything at all and get at least some correlations.

So, Bennett et al. decided to show that you could get correlations just by chance. To make their point their choose an animal everyone can agree would struggle to tell happy and sad people apart, like a fish for example. Then, either because they really wanted to ram home their point that their ‘subject’ was really not thinking about the sad and happy people, or because they couldn’t fit a fish tank in their MRI scanner, they used a dead fish.