Angel #1: Wait, you're telling me that almost 300 years after Laplace and Bayes explained how to do statistical inference correctly, the primitives are still using frequentist methodology? A methodology that is fundamentally incapable of telling them what they actually want to know, which is how likely it is that the effect is real?
Super store Darren, until perhaps the very end. The need for observational data at the end could be questioned (as opposed to having the RCTs in the first place) and the bias in observational data from even estimating the sample average treatment effect will carry over to all “subgroup” effects. Also, just to make the story more complex :-) it could be mentioned that the RCT data alone can often be used to estimate how patients’ treatment effects (on an absolute scale) can be estimated as a function of patient characteristics in the RCT even in the absence of HTE.
Angel #1: Wait, you're telling me that almost 300 years after Laplace and Bayes explained how to do statistical inference correctly, the primitives are still using frequentist methodology? A methodology that is fundamentally incapable of telling them what they actually want to know, which is how likely it is that the effect is real?
Angel #2: Yes.
Angel #1: (weeps softly)
Super store Darren, until perhaps the very end. The need for observational data at the end could be questioned (as opposed to having the RCTs in the first place) and the bias in observational data from even estimating the sample average treatment effect will carry over to all “subgroup” effects. Also, just to make the story more complex :-) it could be mentioned that the RCT data alone can often be used to estimate how patients’ treatment effects (on an absolute scale) can be estimated as a function of patient characteristics in the RCT even in the absence of HTE.
Many thanks, and thanks for the comment Frank.
You’re welcome and sorry for the typo. Super store -> Super story