Hi @danielf : Strictly speaking, the margin of difference is not something that is calculated; it defines a difference that is considered, from a practical/scientific point of view (not statistical), acceptable. And it will depend on the application. That said, it can't be too small or passing the TOST will be impossible. And, if it is too large, passing TOST will be so easy that it is meaningless. I'd push to come up with a margin (non data driven) that makes sense for your particular application and can be defended. After all, if it is derived based on data, and your process is highly variable, you will be rewarded with a margin so large that passing the TOST may be easy...but it may be meaningless and lead to unacceptable consequences (e.g., patient safety).
That said, there is a fairly readable paper in the link below that may provide some guidance...or it will just raise more questions.
https://pubs.acs.org/doi/epdf/10.1021/ac053390m?ref=article_openPDF
