Create a composite score based on standardized individual measures

[Helen Chang]

Hi,

I would like to create a composite/overall measure based on some individual proxies (eg. A, B, C, D, E). Besides ranking these individual variables before adding them together, can I also try to standardize these variables by calculating the z score since they are different scales? If yes, how can I (potentially use SmartPLS) create a formative latent variable using the Z scores of these individual measures to create an overall measure? (Lohmoller, J.B. 1989. Latent variable path modeling with partial least squares. Heidelberg, Germany: Physcia-Verlag). Thanks!

[Joseph Coveney]

I don't have answers to either of your questions, but I have a couple of counter-questions: (1) is there a particular reason for choosing formative latent variable, and (2) is there an advantage to using partial least squares to fit the model? I don't have access to the book, and maybe the answers are in there. It might be of some interest to the List to let others know what lies behind your choices.

If the answers are in the negative, then you might want to consider fitting a conventional CFA (or even EFA) model using -sem- or -gsem-. You might also consider forgoing standardizing the proxy values inasmuch as the factor loadings will take care of scale differences for you.