>> There is a good reason why scientists want something to occur with 90%+ certainty rather than, say, 67% certainty... the latter threshold would basically ensure that progress slows to a crawl. <<
Most scientists would also agree that progress would slow to a crawl if every experiment had to reach 99.999% certainty , because costs , time , and manpower requirements would be prohibitive. There has to be some judgement used , depending on circumstances.
Consider an analogy between drug approvals and skydiving , at two extremes.
Anyone looking to buy a parachute for sport skydiving ( for fun , not because he has a death wish ! ) would expect that parachute to have a near zero chance of failure of the main chute and a near zero chance of failure of the emergency chute. Combined , the overall chance of failure is essentially zero for a properly packed chute and the expectation of this is perfectly reasonable and prudent. This is analogous to the approval standards that should be required to approve a drug to grow peach fuzz on bald men's heads , for example. There should be two , independent , sizable , multi-center , placebo-controlled trials that hit a p <.05 , for a combined false-positive probability of 1/1600 , or less when you consider supporting earlier studies. This is also perfectly reasonable , IMO.
Now , at the other extreme , consider the 'skydiver' who launched himself off of one of the burning WTC towers on 9/11 , grasping a sheet of plywood in hopes of gliding to safety. He probably knew the odds of success were near zero. He would have preferred a reliable parachute , or for that matter an unreliable parachute , or maybe even a sturdy beach umbrella , but the plywood is what he had access to and he felt it gave him some incremental, greater-than-zero chance of surviving , so he gave it a try , understandably. This is analogous to patients with grim , late-stage cancers. They'd like to have a reliable , 1/1600 false-positive-type drug to try , but if there are none , they want the 1/200 type , or the 1/10 type , if that's the best available.
I'm all for standards , and for the need to do good science. I'm also for taking a big-picture approach when it comes to the approval of drugs for deadly diseases where there are currently no effective treatments. There is a greater-good principle involved , IMO , that says it's OK to have an approval process that allows one ineffective drug to be used for a few years before it's recalled , if that same process results in 5 , 10 , or 20 drugs that do provide benefit to get on the market for the extra few years required to conclusively demonstrate that benefit. This , I believe , was the thinking that resulted in subpart H approvals. The failure was in not extending the logic to cover situations like Provenge , which don't involve surrogates.
Market Data 
Markets 
AI
