In a hilarious demonstration of why proper experimental design (and specifically data analysis) is so important, researchers managed to "show" (in a recent conference poster) that a dead salmon responds to human emotion.
Obviously, there was no such thing actually going on.
The correct conclusion? That if a study isn't properly designed (and I include analysis as part of the design), it's quite possible for it to find effects that aren't really there.
We see this sort of thing all the time in the field of autism research... and no small part of it is due to groups like DAN. Gyah.