>i bet if you bothered to look at a distribution of drug performance in ph 3 compared to ph 2, that distribution is skewed bimodally, with a peak at no efficacy (for those drugs that made it through to ph 3 by chance) and a peak at equal efficacy, where i would bet you would find a preponderance of drugs with established MOAs, hard endpoints met in ph 2<
I would expect such a graph to be much more uniform in shape than you postulate—not a “twin peaks” look.
The typical phase-3 trial has plenty of statistical noise (although less than the typical phase-2); hence, whatever the true efficacies of the drugs being plotted, the observed efficacies will be scattered.
Moreover, the true efficacy of drugs that advance from phase-2 to phase-3 is not either zero or the phase-2 observed efficacy—it can be any value in between (or even higher than what was observed). Bad drugs from a commercial standpoint often have weak efficacy rather than no efficacy.
Further, the results of phase-3 trials are, in general, less amenable to sponsor- and investigator-generated bias than are the results of phase-2 trials. This compounds the effect of program-survival bias and creates even more phase-3 failures proportionally than would otherwise be the case.
Market Data 
Markets 
AI
