District internet gap and convergence

The gap is the difference, in working-hours median latency, between a district and the best-served district in its country with sufficient measurement — the frontier, recomputed every quarter. Ranks run within country and within each region; a district is ranked only when at least half of the quarter’s hours were observed and the mean number of responding addresses clears the reliability gate.

Each district’s gap over the last eight quarters is fitted with a trend whose standard error allows for serial correlation. A falling gap with a clear slope is converging; a rising one is diverging; the rest are stagnant. A frontier that improves faster than the typical district shows up as widespread divergence even while every district improves.

Two covariates separate a reach gap from a congestion gap: the settlement class of the district (city, town, rural, from GHS population grids) and the evening spread, the evening-peak median minus the night-trough median.

226,493 of the 476,543 district-quarters, and 8,367 of the 19,128 districts in the latest layer, are low-confidence: their gaps and ranks are NULL while their levels stand.

The frontier can change district between quarters, so a country can read as diverging while its own latency fell; the convergence class is about the gap, not the level.

This is not a last-mile speed ranking (the benchmark story explains why speed tests and this measurement rank districts differently), the package ships identifiers and attributes but no geometry, and the covariate columns are NULL where a district’s hexes could not be assigned.