Mediation muddle

Michael_Jasper

Does the following scenario surprise anyone? It is my situation, which I am a bit perplexed by. 

I have a situation where Predictor 1 is shown to have an indirect effect, via Mediator, on dependent variable. 

And when tested seperately, Predictor 2 is shown to have an indirect effect, via the same Mediator, on the same dependent variable.

But when Predictor 1 is multiplied by Predictor 2, to create a new Predictor variable, we will call Predictor X. This is NOT observed to have an indirect effect. This runs against my expectation.  

Moreover, when Predictor 1 and Predictor2 are tested Together as predictors - there is no indirect effect observed for either.  

Is it just that my expectations are ill placed? 

andersony3k

It's not so surprising to me. To what degree do your data meet all of the underlying assumptions? For example, nonlinear relationships between the predictors and the criterion variable could really screw things up. Obviously, the ideal way to identify mediators would be to conduct experiments. In that way, the causal relationships would be established by means of experimental manipulation.