I’m relatively new in the PLS-SEM world, and, of course, I’m already having big troubles for me…but for sure small (or no) troubles for experts.
I’ve already specified my structural model, it has 3 latent variables. See image below.
Due to the quality of the data available (a quality that cannot anyway be improved), I’ve only dichotomous items for the measurement model and a few continuous ones. I’ve 27 items, 9 for each latent variable. Sample size is 90.
I don’t know the best option for dealing with the data and I’m going to try to explain it with an example. Let’s suppose we are dealing with just 1 of the 3 latent variables (although the situation applies also to the other 2), measured by 9 items: 7 dichotomous and 2 continuous. The options I have are:
1) Measure the latent variable with the 9 items directly. See image below. The problem: I don’t know exactly if it is possible mixing continuous and dichotomous data when measuring a latent variable.
2) Measure the latent variable with 3 items resuming the 9 items. See image below. The problem: I don’t know the best option for resuming the 9 indicators in 3 ones, accounting that we are mixing both continuous and dichotomous items…composite indicators, multidimensional scaling…?
3) Forget about it: mixing dichotomous and continuous indicators is not desirable.
4) Another option I don’t know about…but maybe you do
For working with PLS method I’m reading A primer on PLS-SEM (Hair, Hult, Ringle and Sarstedt, 2013), but few lines are dedicated to the problematic of data metrics and nothing the possibility of mixing categorical and dichotomous data (although I’m still reading it, perhaps it will appear coming soon).
Thanks and good luck