the model that I am currently working on consists mostly of variables with reflective indicators. Yet, there are two variables with formative indicators in the model. Different from most other formative modeled latent variables, I directly asked the informants how important each of the indicators is for them. (Hence, I have the empirical weights for each indicator!)
Take for example the variable project performance:
- How was the project performance on each of the following dimensions?
How important was each dimensions for evaluating project performance?
(1) Since I have the empirical weights for each dimension coming from the questionnaire, I could just use these weights to form an overall weighted index for the variable. In this case I would attach a single-item-index to the variable in the PLS-Model.
(2) Just like in alt. (1) I could use the empirical weights, but instead of forming an overall index, I could attach each indicator multiplied by its empirical weight to the latent variable. PLS would estimate the weights of the indicators (again?!?).
Personally, I think that alternative (1) is the way to go. The thing that bothers me in alternative (2) is that PLS will estimate the weights of the already! weighted indicators.
Could anyone give me his opinion on that issue???
Thanks a lot,
Chris