I am trying to understand what is the better/general way to set the limits of the Region Of Practical Equivalence (ROPE). ROPE would allow for investigating if there is a null effect in the parameters of our model (and combination thereof). A recent paper by Kruschke expands on the subject (ref).
I am aware there is no general rule to set the ROPE limits (it depends on the context and the application). Kruschke, for example, suggests to use half of the Cohen’s small effect size for the ROPE (i.e., \pm0.1) when comparing two groups.
What is your strategy? Do you have any suggestions? In my case, I am dealing with hierarchical models (where the likelihood is not normal, for example I have cases with lognormal distributions, beta distributions, truncated distributions…) and I am wondering if I can apply Kruschke suggestions anyway.