Coding and AI classes in Cannock

Using the Nomis API, the learner describes all 296 English districts by the same 18 numbers: what fraction of residents falls in each five-year age band at the 2021 census. On that measure, the single districts closest to Cannock Chase are Barnsley, Wakefield, Ashfield, Wigan and Bolsover, none of them in the West Midlands. Hierarchical clustering then goes further than nearest neighbours: it builds a complete family tree, merging the two most similar groups again and again until only one is left.

The linkage decides what "most similar groups" means. Single linkage uses the closest pair of members, complete linkage the furthest pair, average linkage the average of all pairs, and Ward's method merges whichever pair of groups adds least spread. Cutting each tree into four families gives startlingly different results.

Single linkage suffers from chaining: districts join the big group one after another through a chain of close neighbours, so almost everything ends up in one family and the only other "families" are three districts with unusual age profiles, the Isles of Scilly, the City of London and Tower Hamlets. That is technically correct and practically useless. Ward's method gives families of comparable size, one reason it is a common choice for this kind of task. With Ward and eight families, Cannock Chase sits in a family of 92 districts drawn from every English region outside London, including 23 from the South East and 16 from the East Midlands.

There is also a way to score how faithfully each tree reflects the original distances, the cophenetic correlation: 0.797 for average linkage, 0.714 for single, 0.696 for Ward and 0.562 for complete. The tree that looks most useful is not the one that scores highest. Choosing a clustering method means deciding what the groups are for, not just reading off a number.

Understanding how AI decides what counts as similar gives Cannock teenagers an advantage with every tool they use, a strong reason to learn to code in 2026. The longer argument is in why teenagers should still learn to code in 2026.