Hi everyone,
I have a question about the difference between structural equation modeling and mixed effects models.
What I'm trying to do is to use anatomical data (e.g. volume of a certain brain region) to explain performance in a cognitive test while controlling for age, sex, education level etc. I want to know how strongly the brain region volume influences performance and ideally to make an inference about causality by controlling for other factors. SEM and mixed effect models seemed like appropriate methods to me but I don't see how they would give me different information.
I'd appreciate if someone could explain this, or recommend any reading about the comparison of these methods, or maybe you want to suggest a different method altogether (e.g. to find out how much of the variance of cognitive performance is explained by that brain region volume).
To be more precise in case it matters: there are 5 different cognitive tests and 5 different anatomical measures, both sets have been measured twice (different years). I don't know if there is a good way to incorporate everything into one analysis or if I need to do 5x5x2 analyses. Thank you in advance for any help!
My guess would be mixed effects but that is because your data seems more like a designed experiment than anything else. Suggest you start with looking up mixed effects models follow up with SEM, especially path analysis. If none of these methods seem to work return to your research questions for guidance. BTW what are your research questions? Best wishes, David Booth
Thank you for your answer! The question is how a specific area of the brain influences performance in cognitive tests that inform about specific cognitive aspects (e.g. memory, executive function). It's a bit more complicated than that but I'm still wrapping my head around it, so the first thing I wanted to consider was which analysis would be most informative.
I hypothesize that smaller volume correlates with worse performance, for instance, but there are several other outcome measures apart from volume where I don't necessarily have a specific hypothesis about their relation to performance.
Hello Tanja Schmidt. Your comments above--e.g., "specific cognitive aspects"--suggest to me that you are interested in 5 univariate questions rather than one multivariate question. In other words, you want to examine each DV separately rather than one latent DV with the 5 cognitive tests as indicators. Is that right? If it is, you'll have to give some thought to how you want to handle the multiplicity problem.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling