Showing the spread, not just the average

Each county's swing minus the national swing, in percentage points. If counties moved independently this would be wide; instead most sit within a few points of the national shift, cycle after cycle.

Each dot is one state: the percentage-point change in prevalence per doubling of tract median income. The dark rule in each row is the median state, the grey rule is zero; on every measure but binge drinking, nearly every state sits left of it.

Self-placement on a seven-point liberal–conservative scale, split by political knowledge. Among low-knowledge respondents the two parties' distributions sit almost on top of each other; among high-knowledge respondents they barely meet.

Each row is the national distribution across all ZIP codes; the gold dot is this one. The wash runs better to worse across each row, the dotted rule is the US value and the solid rule the state's.

Every county in the country, drawn on canvas rather than as SVG — at three thousand marks the node-per-point approach the rest of this gallery uses starts to cost more than it returns. Dot area is proportional to county population, so a county of a million people covers four times the ink of one with 250,000; the smallest counties are held at a minimum visible size, and the key says where that starts. Sizing shows where people are, not a population-weighted estimate. The 31 counties with no 2020 direction on file are drawn in grey rather than dropped.