individually estimated / defined models can be a nice sanity check for the assumptions / complexity of hierarchical models. But in science we mostly want to generalise, so if the hierarchical model is worse in some way, it's probably just badly spec'd, imho...
