Live public record · May 29, 2026
American Policy Atlas
A teaching observatory for statistics built on every U.S. county. Scrollytelling modules walk from a single map to spatial autocorrelation, empirical-Bayes shrinkage, and a California redlining flagship, each with the live numbers and a grounded research-assistant panel.
- Status
- Live public record
- Topics
- observatory · statistics · teaching · census · spatial
- Boundary
- The linked artifact documents the current public release. Research findings require their own paper or dataset record.
An atlas that teaches inference rather than just displaying it. The map of 3,144 counties is the entry point; from there the modules build up the machinery a reader needs to argue about the numbers: confidence intervals on the diabetes rate, the income–poverty regression and why the slope moves when you add a control, Moran’s I (I = 0.60, permutation p = 0.001) for spatial dependence, and empirical-Bayes shrinkage for the small-county problem.
The flagship case study is the California redlining story at the census-tract level: HOLC grades against present-day health and credit outcomes, with the associational claims kept honest. A right-drawer research assistant answers questions grounded only in the atlas’s own corpus, using client-side TF-IDF retrieval on bare GitHub Pages, with an optional Cloudflare Workers AI layer for generated answers when an endpoint is configured. No API key ever ships to the browser.