What are the core assumptions when doing mediation analysis in structural equation modelling? I am estimating the classic mediation model using SEM in panel data in Stata with fixed effects (image attached). Having found a relationship between X and Y in my earlier fixed effects regressions, I want to look at the importance of mediator variables in my data in explaining this relationship, i.e. does information on net worth mediate the relationship I observed between unemployment shocks and mental health scores. I know how to do this in my statistical program, but first I would like to know what the core assumptions (implicit or explicit) are in this kind of an analysis.