HCM-Using lower order components as individual mediator?

[leexiuxi]

Hello all,

I just finished to read Hair et al.'s (2017) book, hoped that I could find the solution of my analysis.

[About my model]
1) Two exogenous variables with reflective measurement model (Involvement/Motivation) that each have three sub-dimensions.
2) Mediator is "perceived values toward tourism destination attributes" with formative measurement model that have five sub-dimensions.
3) Endogenous (exogenous) variables are satisfaction and intention to revisit (Satisfaction-->Intention to revisit).
(Involvement/Motivation-->Destination attributes-->Satisfaction-->Destination loyalty)
I have around 200 samples.

The issue of my model was that my model is overly complex, since each exogenous variables and mediator have sub-dimensions.
As a solution, I decided to use Hierarchical Component Models (HCM), since indicators for each sub-dimensions are highly correlated but, among lower order components (LOCs), their correaltions are low (theoretically, these LOCs are also independent each other). So, basically, I decided to use Reflective-Formative type for mediator, while exogenous variables will be Reflective-Reflective type.
By doing this, I could maintain parsimonious of the model.

But, by doing this, I got another issue.
I also want to identify which destination attributes are mostly explained by Motivation and Involvement sub-dimensions.
That is, which motivation/involvement sub-dimensions are most strongly connected with destination attributes sub-dimensions.

So, this is my main question. "Is there any way that I can use these LOCs as independent factor that are explained by two exogenous variables (or sub-dimensions of each exogenous variables), while I am still using HCM?"
Based on my understnaindg, it might be hard to do it, so I have thought about two solutions, 1) Just focus on structural model, and give up to see the relationship between exogenous variable sub-dimensions and LOCs; 2) Run two different model based on two exogenous variables (Involvement model/Motivation model) to reduce the complexity of model.
I just want to ask to everybody, whether there is any other solution...

Thanks,

Jen

[cringle]

Sounds really complex indeed. We would not use the 2nd-order model approach to make your model look simpler order nicer. If you can, just use the lower-order constructs. PLS usually has no complexity issues. But if you go for the analysis you described, usually the two-stage approach is a good solution. But then you cannot link any other construct on the first order dimension. Alternatively, try this approach here:

Becker, J.-M., Klein, K., & Wetzels, M. 2012. Hierarchical Latent Variable Models in PLS-SEM: Guidelines for Using Reflective-Formative Type Models. Long Range Planning, 45(5-6): 359-394.

Kind regards
Christian