forum.smartpls.com

I'm designing an experiment that has a 2x2 design and a few mediation paths to a single final DV. My study is roughly in the area of information systems. I've seen Bagozzi, R. P., Yi, Y., & Singh, S. P. (1991) and Streukens, S., Wetzels, M., Daryanto, A., & De Ruyter, K. (2010) on analyzing factorial experiments with PLS-SEM. I believe the 2x2 is all within the same causal model.

Is there anyone out there running their experiment condition dummy variables as multiple components of a formative latent construct? It seems like it might take better advantage of the PLS algorithm. I could treat my experiment as a latent construct and interpret the path loading and significance of the formative components, rather than having 3 paths to every other construct for the dummy variables.

I generated some test data to compare and it seems promising, but I'm not really interested in developing new methodology as an aspect of my project.

Any thoughts?

I would rather use the two experimental design dummies as single indicator constructs. That should work well as it is basically the same as using the design dummies in a regression. I have a study in the making that uses that approach, but it is not yet published.