GradStudent posted on Sunday, February 06, 2011 - 8:01 pm
I have 2 measures -- one of anxiety (scored 0 or 1), and one for depression (scored 0, 1, 0r 2).
I would like to test a theory called the tripartite model of anxiety and depression whereby there are 3 factors underlying these constructs.
Given that one measure is score (0, 1) and the other is scored 0, 1, 2, can I pool the items from the two measures and load them on the respective factors as described in the literature? Or do I need to use a specialized technique given that the two measures have different scoring?
I have a question similar to the one posed in the original post. I am planning to replicate a CFA, and eventually conduct measurement invariance testing, on a model that has three latent variables, each with 4-6 items as indicators. Of the 14 items total, 13 are measured on a 6-point Likert scale and one item is measured on a 5-point Likert scale. I was planning use raw data as input and use the robust maximum likelihood estimator.
I am wondering whether I need to treat the 5-point item differently in any way (for example, to use a transformation). I am not only concerned about the CFA itself but am also thinking ahead to measurement invariance testing and how that item may affect multi-group analyses.
You do not need to treat the 5-point item differently. You say you are using the robust ML estimator. This makes me think you are going to treat the variables as if they are continuous. If the items have floor or ceiling effects, I think you are better off treating them as categorical.