Part of my research involves 10 questions (answered on a 7-point Likert scale) to categorize respondents into "Hi" or "Lo" on Supply Uncertainty (in this case). I did some EFA on the 10 questions and they form 3 unidimensional factors (with measurement item Q1 being dropped in the final design).
I have my measurement items arranged in a formative fashion (i.e. "feeding into") my three first order factors, each of which forms the main second order latent variable "Supp Uncer." Alas -- it turns red, as "this latent variable has no indicators."
That's fine -- I need an indicator to complete the model. The problem is I don't have one; there is no "dependent variable" or "outcome" that I'm looking at. To get SmartPLS to solve the path model, I came up with a "bogus" variable, a simple sum of all of the measurement items. It has a regression weight of 1.000 (which is what I expect), but more importantly, the model is solvable and I have regression weights for the other paths.
My questions are two-fold:
1) is there a way to solve the model in the first diagram without introducing this bogus variable?
2) any other issues to be concerned with? (i.e. does it drastically throw off other calculations, etc?) I see that AVE = 1.00 as is Cronbach's alpha, but are other bits (e.g. R^2) correct?
Thanks for your time and input, any and all info / pointing to resources greatly appreciated.
-- Trent