
Message/Author 


Hi there, I have a problem that concerns mediation in SEM. I have 4 predictor, 1 mediator and 1 criterion variable(s). The direct effect for a partcular PV and the CV is positve (as hypothesised). However, when the mediator is entered into the model and the indirect effects are stipulated alongside the direct effects, the direct effect of this particular PV (to the CV) becomes large and inverse (conceptually unlikely). When indirect and direct paths are entered in the model there is evidence of suppression (a beta above 1). could this be the problem? Otherwise, what else would be causing this? Many thanks, T 


Please send the two outputs and your license number to support@statmodel.com. 

Melvin C Y posted on Thursday, January 17, 2013  8:54 am



I am testing a few predictors (15 point scale) on an outcome variable (in SD units). Although the correlations are all very high, above .70, with one pair at .90, Mplus did not produce any errors and there were no negative residuals. Tech4 did not show any correlations above 1.0. As expected the standardized estimates were very high, between .7 to .95. However, the unstandardized solutions were much larger, with one path at 3.0 (in SD units). I ran the CI and it ranged from 2 to 6. Is this possible or due to suppression? 


If the variance of the outcome is large, a path coefficient can be as large as you report. The size depends on the variance of the outcome and the predictor. 

Back to top 

