Dear SmartPLS developers:
I've got a question regarding the pre-segmentation option in PLS-POS. The following information is given within the notes:
"If this option is selected, the algorithm will perform a pre-segmentation in the first round that assigns all units to its best fitting group according to the distance measure.
It will not be checked whether this improves the objective criterion."
What does this exactly mean? Does it mean that some kind of cluster analysis (I have just heared about distance measures in the case of cluster analyses - thats the reason for my assumption) is performed before the PLS-POS routine starts? I have studied the information given in Becker et al. (2013) inlcuding the mathematics provided in the appendix of this paper. However, I struggle a little bit because the PLS-POS algorithm never results in the same segmentation outcome. There are huge differences in the results when I repead the algorithm serveral times. From my point of view this is caused due to the randomized a priori assignment of the cases before the improvement of the objective criterion within the PLS-POS procedure starts. I guess, the provided pre-segmentation option should lead to more stable results because the starting point in each run should be the same or more consistent. Isn't it? My PLS-POS segmentation results for each repetition are very heterogenous regardless whether I use the pre-segmentation option or not. As such this option does not seem to have any influence on my outcomes.
In my view, the PLS-POS algorithm provides huge potential for my research field but the results are too volatile that I could apply this method for any study. What can I do to ensure the robustness of my PLS-POS results that is necessary to get any study published? Will there be more pre-segmentation options (e.g.,on the basis of FIMIX-PLS outcomes, cluster analysis results, or self selected group variables) and/or the possibility to repead PLS-POS algorithm automatically several times (e.g., 100 or 1000) like it is provided for FIMIX-PLS and then get the results where the R² or objective criterion is optimal for all of these runs? If yes, when will these features be available?
I would be very thankful for your answers.
Becker, J.-M., Rai, A., Ringle, C. M., & Völckner, F. (2013). Discovering unobserved heterogeneity in structural equation models to avert validity threats. MIS Quarterly, 37(3), 665–694.
Frequently asked questions about PLS path modeling.
1 post • Page 1 of 1