Why Brain Scans Are Not Always What They Seem

I am the author of the dead salmon study and even I am getting tired of reading about it in the popular press. The original intent was to provide a salient example as to why researchers should use proper statistics in their studies. In the six years since we originally published the paper the neuroimaging field has gotten much, much better at doing the right thing. Now, the salmon is getting to be a tired cliche used to highlight how fMRI is "flawed". Let it go...

There's an insidious fallacy at play in a lot brain imaging studies. I've heard it termed "neurorealism": the idea that something isn't real unless we can point to an area of the brain that "lights up" when it happens. (And the converse: any behavior we can associate with someone's brain "lighting up" must be intimately linked to that behavior.)

It's tempting to look at the visible differences between, say the fMRI of a psychopath and that of a normal person and say "oh, look, area x is clearly different, so we've found the center of psycopathy!", but there are waaaaay too many variables in play for that to be a valid inference.

The best neuroscientists know this is rubbish... but the press does not.