group level paths in S-GIMME/CS-GIMME

[Katrina Aberizk]

Hello,

I am writing about a curious finding concerning a group-level path estimate that surfaced differently between S-GIMME and CS-GIMME. I completed S-GIMME, determined that the cluster solution is stable, and learned that the clusters differ significantly on a meaningful variable external to the time series. After, I completed CS-GIMME to learn how the path estimates may surface when I determined the clusters rather than allow participants to self-organize. In CS-GIMME, a contemporaneous estimate surfaced at the group level that had surfaced only in one cluster during S-GIMME. I wondered whether that path is really a group-level path and perhaps had just missed the group-level threshold at .75, so I repeated S-GIMME with the group-level threshold reduced to .70. However, that same contemporaneous path still surfaced for only one of the clusters in S-GIMME.

I am puzzled by this finding and hope that you may be able to advise. To confirm, all participants were included in both S-GIMME and CS-GIMME, so my assumption was that the group-level paths would be the same regardless of exploratory versus confirmatory subgrouping.

Thanks much in advance for your time.

Katrina

[Katie Gates]

Great question. 

I can't say for certain what caused this in this case, but there is one aspect of the algorithm that likely lead to different group-level paths when subgrouping.  If a subgroup-level path is found in all subgroups, these are pushed up to "group-level" when aggregating the results. Basically they are given a black line instead of a green line since everyone ended up having it estimated. 

The subgroup composition in the S-GIMME results could be to blame. For one subgroup, this path wasn't significant for 75% of the people (or even 70%). Does this seem like it could be what's happening? 

Best,

Katie

[Katie Gates]

* to clarify... when I say "The subgrouping composition in S-GIMME results could be to blame" I mean that the composition of the subgroups could be giving rise to why it's not pushed up to be a "group-level" path. It only exists in one subgroup.