Well there are lots of reasons why in theory - but what I suggest is that one first identifies some robust data, and not just anecdotes or freak results, and then develops an explanation if one is necessary and don't generalise from the particular.
But you will know that:)