Single site? [General Statistics]
❝ Previous posts and responses on significance of group effect were focusing on the Group*Treatment factor.
Which is crap. See this presentation and this article.
❝ What is the situation when group term is associated with a p-value of 0.02 …
That’s at least unusual if the the study was performed in multiple groups at one site.
If this was a multi-center study (group = site), it happens more often.
❝ … and the treatment*group is associated with a p-value of more than 0.35.
Fine, if the study was performed at one site. Acc. to the FDA it will allow you to drop this effect and proceed with ‘Model II’. However, this procedure leads to an inflate Type I Error because you have a pre-test (slides 20–21 of the presentation). Simply use ‘Model II’ with the following effects directly:
- subject(group × sequence),
- period(group), and
- group-by-sequence interaction.
PS: I will have a meeting with the FDA (dealing with a multi-site study) in the near future. Will report back how they reacted.
Edit: In April the FDA accepted my arguments and was fine with ‘Model II’ (without the stupid pre-test in ‘Model I’).
Dif-tor heh smusma 🖖🏼 Довге життя Україна!
The quality of responses received is directly proportional to the quality of the question asked. 🚮
- Group Effect alghazam 2021-01-18 22:32
- Single site?Helmut 2021-01-19 11:32