If I understand what poster DewDiligence is saying, selection bias (program-survival bias) is observed because companies can cherry-pick results to present from Phase 2 trials but it cannot do the same with subsequent Phase 3 trials?
I am new to biotechs and trying to understand how that might work.
It seems that, in general, a Phase 3 trial would be designed based on all that was learned from the previous Phase 2 trials. In that case, the probability of success in the Phase 3 trial would be conditional on all information from the Phase 2 trials. That conditional probability of success should in principle not be less than the a-priori probability of success of the Phase 3 trial (as if the Phase 2 trials never took place). If the conditional and a priori probabilities were equal, (that is, there was no information from the Phase 2 trials to guide the Phase 3 design), then you get an extreme case of selection bias. If there was some learning from the Phase 2 results, then shouldn't `program-survival' bias be mitigated?
Of course these probabilities are all less than one and no one expects ``success' in Phase 3 even if there was``success' in Phase 2 (see COLY). [ I think that is what poster Ocyan is saying?]
In the example of COLY and PFE, may be COLY did cherry-pick its Phase 2 results. But I thought that was what doing due diligence was all about. I would have thought that a big company like PFE would have analyzed all data available, not just results publicly presented, before putting up big money for the Phase 3 trial. Shouldn't that mitigate selection bias as well?
Am I thinking about this wrong? Should I worry about this bias more than I currently do?
Market Data 
Markets 
AI
