The crock is extremely poor experiment design. Automatic F in an experimental psych class, or the stat classes I've taken, or quant chem analysis class, or I'd assume pretty much any hard science. So some random and completely uncontrolled combination of viewing violence, artificial light, novelty response to something new (assuming the game was new to them...), adrenaline (duh), finger/hand exercise, skin exposure to plastic, last but by no means least magic video game cooties has an effect on sleep.

There's a subtle difference between anecdote, data, and information... Due to poor experimental design, this is an semi-interesting anecdote, nothing more. Which is too bad.

Edited note: Its the timing thats bad. Doing a huge wide ranging experiment is perfectly valid iff you've already got tons of verified data and info of ALL the constituent parts and can subtract that statistical noise from the huge # of variables experiment. Its like giving higgs boson search raw data to Newton as a first experiment rather than starting with an apple off a tree. For example, medical sample of adrenaline level at bed time fed into formula to subtract out effect of excitation. Then medical experiment using scrambled video signal so the same melatonin effect based on raw light level and average color/brightness can be subtracted out of the sleep data. After subtracting out about 10 correction factors they might have real data, or even info, instead of anecdote.