'The primary factor should be if the voters of the district can easily identify what district they're in.
I agree. That's the advantage of dividing the state into identifiable geographical subunits and then putting the subunits together to make districts. Larger districts like congressional work better if one starts with larger subunits. Smaller districts need smaller subunits to start. I've found that subunits that are no larger than 10% of the ideal district size work best.
This is from the analysis I did on whole county (and whole New England town) maps that were drawn for congressional districts on threads here last cycle. A range of 3.5 (average of 10 geographic units per district) is just under 0.5% of a CD. A range of 2.5 (average of 20 geographic units per district) is just under 0.05% of a CD.
The analysis would generally apply to any collection of subdivisions, not just counties. For example one could use school districts in populous suburban areas as better indicators of communities of interest that are easily identified by voters.
I’m curious how counties play a role in states like Georgia, North Carolina, Texas, New York, and California when trying to run broad analysis like this. These states run the gambit of county size with some having counties that could have a dozen districts, or districts that could have a dozen counties.
I think a lot of this has to do the fact I come from a fuzzy area between major COIs so I’m especially interested as what to do with the buffer regions. I’d be interested to see a map of how you’d draw subsets in a given state so I could get a better idea of the concept if you happen to have a map handy
For the study used to make the graph I used plans that nested as many whole districts in a county as possible and attached the remainder to other counties. That's why the graph refers to regions, since some reflect multiple districts due to large counties. It doesn't imply that nesting districts is best, but nesting them made a uniform approach to handling large counties across the states.
There are lots of threads from 2012-2015 that looked at rational ways to subdivide large counties in different states. Most are available through the old 2010 redistricting directory thread. IIRC MI, WA, and VA got a lot of attention on big county subdivisions.
One important consideration we found based on the plans drawn and the observations of real legislative gerrymanders is that any geographic subdivisions have to be specified in advance and agreed to by all sides. It was too easy to construct gerrymanders (sometimes subtle) and then define the CoIs after the fact. Agreeing on subdivisions in advance block that sort of gerrymander.