when you have a small, homogeneous state like kansas or new brunswick, the assumptions in the modelling should work out fine for whatever kind of outcome you want.

but, when you have a big, heterogeneous region like ontario - which is at least four very distinct regions with totally different everything - all you're doing by aggregating and modelling is polluting sample.

data from kenora or sarnia is worse than completely useless in modelling brampton, it's actually distorting and misleading.

you can do this right using distinct and mutually exclusive sampling frames, but nobody does it. and, how accurate you are depends on how precise the frames are.