Hi,
I have a couple of questions concerning the calculation of the contribution of explanatory variables to the prediction of a dependent one :
For Tenenhaus et al. (2005), this contribution can be estimated by the following formula : sum of Beta (path coef) * correlation but this make sense only when beta and corr have the same sign. My question are :
1) Is it possible to consider that all sign are positive and to use the asolute value of these parameters when all sins are not identical ?
2) Is it possible, when the model integrates mediating variable, to integrate the indirect effet? Imagine, for exemple, a model with 3 variables A --> B --> C. From the preceding formula, it is only possible to estimate the contribution of A on B and of B on C. But A has an indirect effect on C (total effect - direct effet of A on B and of B on C). If I understand correctly, this indirect effect is calculated by this formula direct effect of A on B * direct effect od B on C. So, it seems possible to take into account the contribution of A on C (even if no path is drawn between them) by estmating the contribution of this indirect effect, isn't it?
I hope my explanations are clear and that someone could help me.
Regards.
David.