News for hackers. Stuff that matters.
Re:Placebos becoming too effective, pharma industry worries by richardjordan on Thursday March 28, 2013 @11:12AM (#5455252)
What is problematic is when you set out to develop a statistical approach purely designed to increase the rates at which experiments deliver positive outcomes. You start letting the tail wag the dog when you start from the assumption you are testing stuff which must have a significant benefit, and ask how you can better demonstrate that benefit - rather than accepting that the results appear to show that the benefit is not particularly statistically significant.

Drug trials are already done in circumstances incredibly favorable to the drug companies. If even in such a situation a potential drug is struggling to show benefits the solution isn't to keep changing the way you're testing until you get the result you seek.

Re:Placebos becoming too effective, pharma industry worries by uvdiv on Thursday March 28, 2013 @12:40PM (#5455905)
You start letting the tail wag the dog when you start from the assumption you are testing stuff which must have a significant benefit, and ask how you can better demonstrate that benefit - rather than accepting that the results appear to show that the benefit is not particularly statistically significant.

There is no such bias in this method. It could just as easily show a drug to be worse than a placebo (harmful) than better: there is no prior assumption in either direction. The method merely increases the accuracy of this result.

The fact that it increases the number of positive drug outcomes is a good thing: we want more working drugs. Hiding drug effects with weaker tests, giving more inconclusive results, is not an advantage.