Skip to content
BroadScout

Methodology

How these figures are produced.

Every number on this site is traceable to a retained source file. Nothing is modelled, interpolated or inferred, and where a source is silent the page says so.

Import, not scraping

Source files are downloaded once from the publishing agency, retained, and imported offline. An import is immutable: it produces a dataset with a fingerprint over its inputs, and re-running it against the same files reproduces the same dataset. Serving never writes. A page can therefore only ever show what some recorded import put there.

Which place a record belongs to

Assigning a record to a city is a decision, and it is versioned separately from the data. City membership comes from the Census Bureau's Block Assignment Files, which state which census blocks fall in which incorporated place. A geography version is marked authoritative only when the official file has been obtained directly from the publisher and its bytes match what was archived; otherwise it stays marked research and the pages say so. A new version may sharpen a label but may never move a record into or out of a place.

Availability figures

Providers file with the FCC which technologies they offer where, and at which advertised downstream and upstream speeds. Those speeds are published here only as the explicit pairs that appear in a filing — a downstream figure is never shown beside an upstream figure from a different record. Residential and business service are distinguished, and only residential service is summarised on a location page.

What a filing asserts is that a provider could offer service to a location within a stated time. It is not a statement that service exists at an address today.

Adoption figures, and their margins

Adoption comes from the American Community Survey, which is a sample survey rather than a count. Every estimate is published with a margin of error at the 90% confidence level, and this site shows it next to the estimate rather than hiding it. Two figures whose margins overlap are not different from one another, however different they look.

A place is published as a page of record only when it has at least 500 households and its broadband margin of error is no greater than half the estimate. That second test is the Census Bureau's own reliability convention expressed as a ratio. Places that fail it still get a page, because the place is real and the figures are the published ones, but the page states each specific reason the estimate is thin instead of presenting a rate.

Where a survey finds no household without a subscription, the page reports exactly that. It is never rendered as "100% of households have broadband", which would turn a property of the sample into a claim about the place.

What is deliberately not done