> I do find the focus on compactness and contiguousness kind of strange.
They are "looks nice" (in a picture of district outlines) criteria that actually (given the geographical distribution of party members) are equivalent to deliberately gerrymandering in favor of Republicans.
And that's really the key point: as long as you have single member districts and data about the existing distribution of voting preferences (which people in a position to adopt or sell policies do), you can design facially neutral criteria whose application will result in any distortions of representation you could seek to gain from gerrymandering, and say "let's automate with these criteria, take people out of the loop, and avoid gerrymandering", when all you are really doing is automating gerrymandering.
>you can design facially neutral criteria whose application will result in any distortions of representation you could seek to gain from gerrymandering
I seriously doubt this assertion is true. If you were to take, say, one state and design a set of district division rules based on, say, population density, geographic similarity and boundaries, infrastructure boundaries, existing political borders like counties and cites and manipulate the weights given to each factor so that one party comes out over- represented, the very next time the districts are redrawn after significant changes that carefully constructed balance will fall apart.
> If you were to take, say, one state and design a set of district division rules based on, say, population density, geographic similarity and boundaries, infrastructure boundaries, existing political borders like counties and cites and manipulate the weights given to each factor so that one party comes out over- represented, the very next time the districts are redrawn after significant changes that carefully constructed balance will fall apart.
No, there's not. The patterns of geographical distribution of partisan identification in the US is relatively stable over decades (and that of broad ideology is even more stable, as much of the shift in distribution of party identification is realignment of parties rather than redistribution of ideology), it's also pretty similar across the nation in many key ways.
I started hearing about and researching the idea of algorithmic districting to address gerrymandering the same year I started college, 27 years ago; optimizing for compactness and contiguity was a popular proposed standard then, and that it overrepresentingamong conservative views and the Republican party because urban cores nationally tend to contain liberal supermajorities whereas rural and suburban areas broadly have a narrower conservative majorities was true and not even a new observation then. And it remains just as true now.
They are "looks nice" (in a picture of district outlines) criteria that actually (given the geographical distribution of party members) are equivalent to deliberately gerrymandering in favor of Republicans.
And that's really the key point: as long as you have single member districts and data about the existing distribution of voting preferences (which people in a position to adopt or sell policies do), you can design facially neutral criteria whose application will result in any distortions of representation you could seek to gain from gerrymandering, and say "let's automate with these criteria, take people out of the loop, and avoid gerrymandering", when all you are really doing is automating gerrymandering.