Comparing diagnostic groups on results of CS-GIMME

[Laura Berner]

We have results of a confirmatory subgrouping GIMME that indicate that one of our groups has three more paths to a particular node than the other group, which has zero paths to that node. There are two paths that the groups share that involve that node, but the groups don't differ on connectivity strength of those shared connections. 

We're wondering if there's a way to follow up on the descriptive finding that one subgroup has a lot of connections to a node that the other doesn't. We're stumped because the individual graphs for each participant within each group will be forced to have edges that exist at the subgroup level and will be forced not to have edges that don't exist at the subgroup level.

Would it make sense to run indSEM on everyone separately (so we're not giving the model any constraining information at the group level), and then compare our pre-defined subgroups on the degree of that node (since we'd then have an individual-level degree metric for each subject)? This would result in less reliable edges, I know...

Another option would just be to run gimme on everyone together as one group, but then we're still applying a group-level constraint that I'm not sure we want to.

Any other ideas are very welcome, and thank you!

Laura

[Katie Gates]

Hi Laura, 

Just for clarification, it is possible for individuals in a subgroup to have paths that are "subgroup level" for another group. Nothing in the algorithm restricts this from happening. So, if no individual-level paths exist to that node in the subgroup that didn't have the subgroup-level paths, that is not because they were forced to not exist. Does that make sense? 

You could always compare strength for that node between the subgroups by, for each individual, summing all paths that go in and out of it (or just in, if you want in-strength, and out-edges, if you want out-strength).  This should have variability across individuals in each group since it sounds like there are some group-level paths for that node.

hope this helps, please let me know if something isn't clear or won't work for some reason.

Katie