I am using a pathmodel to analyze a multiple mediation. I have three exogenous variables ($x_1,x_2,x_3$) and two mediator endogenous variables ($m_1,m_2$) and one final endogenous variable that is the criteria / dependent variable ($y$).
Since this has not been tested previously there is no theoretical reasoning for constraints. So it is a fully saturated model (with $df = 0$). I am fairly unexperienced when it comes to SEM / Pathmodel, however I know that in a saturated model, the fit statistics are perfect and are probably not supposed to be reported. My first question would be, is ok to report a fully saturated model and just interpret the weights of the paths (and dont even bother to report fit statistics)?
A follow up questions stems from a tipp I recently read on a forum. It is, that you could successively constrain non-significant paths from your model, i.e. delete them from the lavaan model specification. As with this, you increase the degrees of freedom and the model fit statistics become interpretable. However I am not sure if this is a good practice? As it seem arbitrary to me to just cut insignificant paths from the model.
