Hello Smart-PLS-community,
I already posted my questions but unfortunately so far I haven't received any answer. Maybe it helps if I reformulate the problem. I was wondering whether it is possible creating PLS-Models with quasi almost interaction terms. I'm wondering because the constructs in my model shows very insufficient reliabilities.
To create the interaction terms I first set up a different model to measure the reliabilities of all the constructs later used for the interaction terms. Then I applied the standard approach: Formative interaction terms were calculated by multiplying the LV-Scores while reflective interaction terms were set up by multiplying the single indicators. The multiplying was based on the optimized single constructs and I didn't eliminate any indicators of the reflective interaction terms even though there are many indicator combinations showing loadings below 0,70. Finally I created a new model by substituting the elements of the first model by my interaction terms, letting in only 2 of the original not interaction term constructs). So my first questions here:
Do I have to eleminate indicators from the interaction terms that do not show sufficient loadings? Would 0,7 be the right value for cutting of indicators? Can I cut of the elements of the interaction terms or do they need to remain in the model similary to calculation of moderating effects?
My final model now consists of 4 interaction terms + 2 normal exogenous variables.
To make it more comprensible, on the left there are two antecedents both being interaction terms (CA=CPxCS;SA=SIxSP).
In the middle I have one dependent interaction term variable (ExplxExplo; depending on both antecedends) and 2 exogenous reflective variables that are the elements of the interaction term (Expl, Explo). The 2 exogenous variables are not linked to the antecends.
On the right there is one depending interaction term (OPxSP) which has paths to all the 3 variables in the middle.
Now I experienced the following reliability issues. First both exogenous reflective variables in the middle (Expl, Explo) show predominantly negative loadings of their indicators and horrible reliabilities. This changes immediately and reliability as well as loadings becomes strong when integrating additional depending variables on the right (that are not interaction terms). I don't want to integrate these further variables so my next questions:
Do I have to care for the reliabilities of the normal exogenous variables in this model or can I nevertheless trust the results? Is there any alternative to fix this reliability issue?
A part from the low reliabilities of the two normal exogenous variables, one interaction term in the model shows also horrible reliabilities. And this doesn't change even when introducing the additional depending variables as described in the paragraph above. So my question:
Do I have to care for the reliabilities of interaction terms or is it sufficient to take care that the single constructs forming the interaction terms fulfill the reliability criteria?
I'm aware of the quantity and the complexity of my models and my questions. I would really appreciate to here your opinions on the difficulties as I'm stucked and have no clue how to address appropriately these problems. If you feel the need to see the models in order to answer the questions, just let me know and I will mail them to you immediately.
So far Smart-PLS and the whole PLS-Stuff are fantastic, would be great if you could help me to finish my work.
cheers,
Patrick