Wed, 05/12/2010 - 12:27

I understand that Mx could be used to handle ordinal manifest variables but I could never follow the one example script I had.

Is it possible to have ordinal manifest variables using OpenMx? I hate to ask seeing how confidence intervals have now been added (yea!) which would allow me to use OpenMx in place of Mx for a number of analyses.

I have 3 raters rating 2 eyes for each subject so that there are 6 measurements for each subject, something like:

subject od.1 od.2 od.3 os.1 os.2 os.3

1 X X X X X X

2 X X X X X X

etc.

where od = right eye and os = left eye. Each rating is on an ordinal scale from 1 to 10. I would like to use a measurement error model that has two latent variables (the "true" rating for each eye) and allows for the correlation between the true ratings for each eye (results for both eyes typically tend to be highly correlated in my experience). (I've done similar models where the measurements were quantitative, continuous, and Normally distributed.) I want to use the model to describe the agreement among the raters.