Correlated, not causally related, variables

Frequently asked questions about PLS path modeling.
Post Reply
PLS Junior User
Posts: 3
Joined: Mon Jul 13, 2015 11:48 pm
Real name and title: N Hart, Assistant Prof

Correlated, not causally related, variables

Post by nnhartma » Wed Jul 11, 2018 3:19 pm

I had posted the following in the Application of PLS section but think it is probably better in the FAQ(Methodology section)

One of the benefits of CB-SEM is being able to model a relationship between predictors without specifying the direction of effect via doubled-headed arrows. I understand that we do not have this double-headed option in PLS because of assumptions underlying the approach. But many times, we will have exogenous predictor variables that are theoretically related to one another (e.g. multiple dimensions of climate or job satisfaction) and we would like to recognize this without specifying a direction to the effect. For example, ClimateVariableA, ClimateVariableB, and ClimateVariableC influence Turnover Intention. Such Climate Variables are likely to share moderate correlations (but not be a higher-order factor) but a researcher may not want to specify a single arrow relationship between any of them.

Does SmartPLS account for this shared variance when estimating the model? If not, can we somehow model it or is SmartPLS a flawed analytical approach when we would expect exogenous predictor variables to be theoretically related to one another but do not have theoretical justification for modeling the direction of such impact?

Post Reply