I guess the difference is structure 1 is the standard way of doing it and so it’ll be clearer what’s happening to someone else reading the model.
I guess you have this extra degree of freedom to scale the parameters, but I don’t see any reason to do it.
I wouldn’t read into the variances of grp_raw. grp_eff and grp_sigma should tell you what you need to know.
Explain this more. Do you have numbers? I don’t think I addressed it.