MacKinnon (2012) writes on p11 that
A moderator is a variable that changes the sign or strength of the effect of an independent variable on a dependent variable. It is typically (but not always) an interaction such that the effect of an independent variable depends on the level of the moderator variable.
As far as I can tell MacKinnon doesn't give an example of when this could happen. Presumably it's possible for a moderator to change the sign or strength of a an effect of the IV on the DV but for this to somehow not to depend on the level of the moderator.
Can anyone give a concrete example of when this could happen?
MacKinnon, D. P. (2012). Introduction to statistical mediation analysis. Routledge.