It says that if you tell me how many counties, large counties, and districts in a state, I can tell you how small the range (or deviation) should be in a plan using whole counties. If it isn't about that number or smaller than that, then you probably didn't work hard enough, or were trying to preserve other elements in a plan like low erosity or UCCs (This graph from 2012 was the original basis of my Pareto model of trade offs). It also applies to any other building block used to form districts as long as their populations are distributed like the counties in a typical state. The fascinating thing is that it shows that the states are quite similar in this regard, and New England towns distribute themselves in population like counties in other states.
Ah, it is a median versus mean thing or something that causes most states to be below the regression line. That detail was what confused me. How does the log thing work/the theory behind it?