Latent growth curve modeling
I have a two-wave twin panel design (the same phenotype measured twice, two years apart for both twins) and I am planning on assessing stability. So far I have worked with Cholesky and the results seem interesting. However, I am really interested in biometric latent growth curve modeling. The obvious flaw would be that I only have two waves available. In the non-genetically informative literature LGC modeling is sometimes performed with only two waves, though no trend can be identified with the data. I am wondering, if I am doing a two-wave "biometric" LGC modeling design:
(1) does this make any sense? Or is cholesky (or maybe simplex) the better way?
(2) What are possible issues and limitations (in comparison to three or more waves)?
Thanks a lot in advance!
Two occasions
I think the Cholesky will tell you pretty much everything you want to know about your two-occasion data. LGC for level & slope requires 3 time points. A latent change score model, twinified, might be a more developmentally oriented.
HOWEVER: Be aware of the ages of your participants. If they are all the same age (e.g., measured on 10th and 12th birthdays) then all is well. If, however, there is variation in age at assessment of the twins, then you will be largely modeling the effects of being measured twice, along with some of the changes with age. Sometimes a developmental perspective can be obtained by specifying many - say 20 or more - ages at assessment and treating the data as if most of it is missing with just a couple of observations per person. In this approach, it may be possible to figure out if development is more rapid between one pair of ages and another. On the whole, strive to conduct age-based analyses, unless the effects of retesting are of interest.
Log in or register to post comments