Jon Heron posted on Thursday, January 05, 2012 - 12:12 am
Happy New Year to you!
I am scratching my head over something that should be really simple. I have an exposure derived from a 4-class mixture model using six binary manifest vars. I have a distal outcome U which is also binary and a bunch of potential covariates. It's very similar to "Figure 19.7 GGMM Diagram for LSAY Data" in Bengt's Kaplan chapter, except I've gone LCGA rather than GMM.
I would like to estimate the association between class C and outcome U and then assess attentuation when including a number of confounders.
I have begun to wonder whether this is possible since when U is manifest, the C~U relationship is a loading within the measurement model for C rather than an association within the structural part of the model. Furthermore, as the inclusion of the covariates can impact on the measurement model, there's no guarantee that any attenuation would be due to confounding anyway.