[jcranmer]

The plural of anecdote is not data. Yes, it's a pithy saying, but it's completely true. Scientific validity requires changing one variable, and only one variable, and seeing if there is an observed effect. The real world is noisy, and anything involving biology incredibly so, which means you need larger sample sizes to distinguish signal from noise. We also know that there is some degree of mental control over physiology--see psychosomatic disorders and the placebo effect--but controlling for that is fiendishly difficult when you can't double-blind the study.

Given the baseline potential for high statistical noise, biological studies in general tend to have poor rigor. fMRI, which you've cited, is notorious for giving problematic results (see the dead salmon study).