Post-Hoc Assumption Tests

In Datamethods, model assumptions are often discussed, and what to do when they are not met. It has already been explained on several occasions that rational alternatives include using methods that relax the most common assumptions: e.g., through Bayesian priors, Student distribution, etc.
In datamethods it has been explained in several places, that using assumption tests to adjust the model parameters usually alters the operational properties of the tests and implies making new assumptions, so it is usually not a good idea.
However, in my experience it is still relatively common for some reviewers to question the assumptions of the model once it has been adjusted (e.g., the pH, PO, normality assumption, etc).
I wanted to ask if it is possible to explain with real, tangible, practical examples, not just theoretical reasons, why it is not a good idea to use post-hoc tests of assumptions in some of these cases, and what exceptions there are to this rule.

1 Like
1 Like