Hi,
I could need the help of an expert to assess whether my approach to calculate a rather complex model with smartpls is feasible (and ideal).
Here is the task:
My model consist of four 2nd-Order-Latent Variables (A, B, C, D). A relates to B, C, and D. B relates to C and D. C relates to D. The direction of causality is from A to D. (In other words, B and C are mediators of the relation of A and D with the speciality that also an effect between both mediators is present.)
All 2nd-Order-Latent Variables consist of formative 1st-Order-Latent Variables (at least two). All respective 1st-Order-Latent Variables consist of reflective Manifest Variables (at least two).
Six Control Variables shall be considered with respect to (B, C, and) D. Six Moderating Variables shall be considered with respect to all relations among the 2nd-Order-Latent Variables. The sample size N is 191.
My approach on how to calculate this model consists so far on five steps.
1. Use of smartPLS due to model complexity and sample size.
2. Calculation of the model (as described above) with the Factor-Value Approach in accordance with Wold (1982).
3. Test for significance of Control Variables in relation to D in accordance with Liang et. al. (MISQ, 2007). The significance level of p=0.1 is applied here.
4. Assessment of mediating effects in the model in accordance with Baron et. al. (1986) and Sobel et. al. (1982).
5. Assessment of moderating effects by group comparison (split of the data sample for each Moderating Variable and separate calculation) in accordance with Edwards and Lambert (PsychMeth, 2007). Then, test for significance for each path following Chin (2000).
Remaining questions:
i. Is the approach feasible (and ideal) for the described task?
ii. The “Chin-Rule” (N > number of incoming paths of formative Latent Variables x 10) is applied. Are there any other rules that have to be considered?
iii. Is it valid to select the Control Variables based on the p=0.1 level?
iv. Can I test for the significance of Control Variables not only on D but also on B, and C? How do I do this correctly?
v. Is there an alternative to the Factor-Value Approach that allows for the calculation of the model (that consists of formative 2nd-Order-Latent Variabels) in one step?
vi. Any other comments?
Any help on these issues is very much appreciated.
A gazillion thanks in advance!
With best regards,
Jörg