P values in clinical trial

#1
Dear all,

I read in this article on p-values that the use of p-values is increasingly criticized in the field of clinical trials.
One of the reasons is that the 5% level is not fully meaningful.
I was interested to know if other alternatives have been proposed for clinical trials decisions.
Thanks
 

hlsmith

Less is more. Stay pure. Stay poor.
#4
Yes, the Null Hypothesis Statistical Testing (NHST) paradigm is not great. Also, because it is all or nothing (black or white) decision making and does not directly take into account existing information. Bayesian modeling would be the alternative and is based on posterior distributions (probability of all effect estimates), not a single point estimate.
 

Karabiner

TS Contributor
#5
Yes, the Null Hypothesis Statistical Testing (NHST) paradigm is not great. Also, because it is all or nothing (black or white) decision making and does not directly take into account existing information. Bayesian modeling would be the alternative and is based on posterior distributions (probability of all effect estimates), not a single point estimate.
The NHST analyses are easy to perform, but results seem nearly uninterpretable, or simply useless, with regard to what we are really interested in. In contrast, Bayesian methods require much more sophistication, but results can be interpreted quite straightforward.
 

hlsmith

Less is more. Stay pure. Stay poor.
#6
Also of note, you do not have to perform corrections for multiple comparisons in Bayesian analyses, given you provide the correct prior or a reasonable prior for the comparison. So you can compare any and all combinations of groups - you just need to provide an informative prior.