Allow nesting based on other variables

[emitanaka]

Something like:

design() %>% 
  set_units(group = 2,
                   subject = nested_in(group, 5),
                   cells = nested_in(subject, .based_on = group,
                                                           1 ~ 2,
                                                           2 ~ 4))

[macroscian]

Looks really useful - perhaps an even more general nested_in(subject, ~ifelse(group=="group1",2,4)) where formulae without a lhs would be evaluated to provide the nested levels.

[emitanaka]

I agree with keeping it general like you say. Let me think about how this affects the underlying graph structure (and subsequently the suggested design_model()). Part of the problem is that you can end up specifying a crossed structure but it won't be encoded as such in the system and hence causes issues downstream. So in that sense, we might need some constrain so it doesn't allow users to go too wild.