While my own blogging languishes (it's the end of the year in pharma and biotech in an era where you have to do more with less), I refer you to John Cook's entry on the cult of statistical significance. The only thing I have to add is that I've seen cases in which where Type II errors are more important than Type I errors, I've seen the bar for statistical significane set at 0.1 rather than 0.5. While less conservative than 0.05, it doesn't erase the fundamental problems with statistical equality testing of safety problems.
Maybe this is why I get the feeling for every protocol that passes my desk that study designers are shooting themselves in the foot.
