We group the 21 authorities four ways: overall, and separately by economy, by population and by economic geography. Two authorities can have very similar economies yet differ in almost every other respect, and those gaps often matter more for policy than an overall score.
Select a cluster on the map or in the list. Bars show how far the cluster's average sits from the all-MSA average, in standard deviations.
Straight-line distance in the clustering feature space. Pick a neighbour to see which variables pull the two together or push them apart.
With only 21 areas, closely correlated variables are grouped and one representative kept from each, so the model is not counting the same signal twice. Every variable in it stays a real named measure, so a cluster can be explained in plain terms. Variables marked pinned are kept regardless of grouping because they are policy-salient; those marked via are still reported in the comparison above, just not counted separately in the distance.
Adjusted Rand Index — 1.00 is identical grouping, 0.00 is no better than chance.
Those two views agree at 0.01 — effectively chance. How a place is shaped tells you almost nothing about who lives there. That is the clearest argument for keeping the views separate rather than collapsing them into one index.
Two known gaps, both needing a data fetch rather than a method change: sector composition (the business survey extract has no industry breakdown) and household tenure composition.
A different question from the tabs above. Rather than asking how similar these places are, this groups MSAs by how far devolution has gone in each and how able each is to act on it, to find places facing shared challenges. Read the other way, it also shows which places differ most on a given dimension, which helps when building a more varied set. It uses hand-scored assessments rather than published statistics, and covers the MSAs on the MSA Academy Programme.
Each dimension is rescaled to 0–1 before weighting, so the 0–21 power score does not swamp the 1–4 scales. Set a weight to zero to drop a dimension entirely.
Ranked by weighted readiness score, then cut into groups at the largest gaps in that ranking (Jenks natural breaks) — so a group boundary falls where there is a real step, not at an arbitrary round number. Any group below the minimum is merged into its nearest neighbour, which can leave fewer groups than requested.
Devolved powers on the vertical axis against any measured variable, to see whether capability lines up with economic or structural conditions. Points are coloured by the group above; click one to read it out.
Group on both at once — devolved powers alongside, say, productivity or settlement primacy — to find MSAs that are alike on capability and on the conditions they are applying it to. Everything is standardised before weighting, so scales do not compete.
The scoring is deliberate expert assessment, and the notes below record where it is least certain. Treat the groups as a starting point for discussion rather than a final allocation. They are not a performance ranking and should not be presented as one.
The seven areas of competency in the devolution framework, scored 0–3 each. Two MSAs on the same total can hold quite different powers, which matters when choosing who to put together for a specific session.
