What is it about?
Correlation between the test for carry-over and the test for treatment based on first period values only strongly biases the significance test for treatment if based on the test for carry-over, as was shown by Peter Freeman in 1989. This is explained and illustrated in this viewpoint piece.
Featured Image
Why is it important?
It shows that preliminary testing to guide choice of analysis can be dangerous and that the properties as a whole of such procedures should be analysed.
Perspectives
I was a reviewer of Peter Freeman's article for Statistics in Medicine and had independently come to the conclusion that there was a problem with the procedure. Peter had analysed the issue in depth and it was a very simple matter to recommend publication.
Professor Stephen J Senn
Consultant Statistician
Read the Original
This page is a summary of: Viewpoint
: Do not resurrect the two‐stage procedure, Pharmaceutical Statistics, July 2022, Wiley,
DOI: 10.1002/pst.2224.
You can read the full text:
Resources
Testing Times
A blog explaining the problem with the tow-stage procedure
The AB/BA Cross-over: How to perform the twostage analysis if you can’t be persuaded that you shouldn’t.
This paper, which is otherwise difficult to get hold of, describes how the two-stage procedure can be altered to maintain the nominal error rate.
Contributors
The following have contributed to this page