Thread by @xenocryptsite

Yeah again I think this is one reason the 2016/2020 swing map is more confusing, because the correlation is already apparently high.

https://twitter.com/Nate_Cohn/status/1330134386818347013

From the 2012/2016 swing map, we're trained to look for big trends, but by definition if things are mostly stable with high correlation, you're not going to find those.

Mostly stable, that is, at a county level. There can be swings that are harder to see at a county level, like among gender, age, and actually, Hispanic population outside of Florida and the Texas border.

(Not to the same extent, but that's where a lot of the outright majority Hispanic counties are.)

That's a bit imprecisely phrased. What I mean is, maps can only even hope to inform you about a subpopulation to the extent that subpopulation is unequally-distributed over the geographic unit of the map, and even then, there's all the risks of Simpson's Paradox etc etc.

The 2016 county swing map is so striking because there was enough swing among non-college White voters, AND because there was enough concentration/segregation of that group at a county level. You need both.

Now I don't have the uh, county-level Gini coefficient of non-college Whites vs. the county-level Gini coefficient of Hispanics, so I could be wrong.

(Sorry, I initially used a pre-election map from whatshisname, Ruffini, just was trying to find a labeled one.)

Of course to avoid Simpson's Paradox you need to avoid any swings from other groups correlated with the presence of your group, which might also make this Hispanic swing stand out less in some cases.