A Census of the World's Local-Government Websites

A global census of 755,989 distinct sub-national administrative units across 185 ISO-3166 territories, of which 156,287 rows carry an official-website address in three provenance layers with measured per-layer precision. Each unit is keyed by a stable jurisdiction identifier, crosswalked to national statistical codes, Wikidata and GeoNames, and enriched with resident population, settlement class, machine-legibility and transport-security signals, hosting and operator jurisdiction, sovereignty context, and site-emergence dates. Version 1.3.0 is a correction release. The three provenance layers now ship as a column, provenance_layer, rather than existing only in the descriptor's prose. A second column, tier_in_frame, marks the 138,808 units that sit at a country's governing local tier; 182 rows that are country relations or macro-regions were removed from that frame. Population must be summed only where frame_population_basis equals governing_tier, because the frame nests tiers and the naive sum reaches 1.5 times the world population. A four-value presence_state supersedes the letter-graded absence_grade, which is preserved as absence_grade_legacy. An integrity audit of held addresses removed 10 whose domains had passed out of government control, replaced 7, and flagged 18 that are still governmental and serving injected search spam; the post-stratified suspect rate across held addresses is 1.101 percent [0.642, 1.560]. Cross-border duplicate flags introduced in 1.2.1 are retained: rows that are second copies of a unit present under a neighbouring country are marked relation_duplicated, with one copy per relation marked relation_primary. Analyses keyed on country should filter to relation_primary, and analyses of local government should additionally filter to tier_in_frame equal to in_frame. Read USAGE_CAVEATS.docx before modeling: an empty cell records that a signal was not measured, not that it is zero. Any absence is additionally conditional on reachability from a United States vantage, which ranges from 38 to 98 percent by country on addresses already verified live. Computation used the Holland Computing Center of the University of Nebraska, which receives support from the Nebraska Research Initiative.