Change Ringing Corpus · Roadmap item 23 · A negative result

Everyone says there are
two kinds of ringer.
There is one.

What two populations would look like

Ringers describe two worlds. The Sunday band turns up week after week at one tower. The peal circuit is smaller and more mobile, travelling to ring long performances. The received wisdom is that these are largely different people who happen to share a set of buildings.

If that were true, the share of a ringer's performances that are peals would be bimodal — a hump near zero for the Sunday band, a second hump near the top for the peal circuit, and few people in between. That is a shape, and a shape is checkable.

If the folk model were right — illustrative, not data

Drawn by hand to show the shape being tested. This is not a measurement and no number in it comes from the corpus.

What the corpus actually shows

One mode with a long thin tail. The median active ringer spends of their reported ringing on peals, and ringers who ring only peals essentially do not exist.

It is not two kinds of tower either

The same test, run over towers rather than people: what share of a tower's performances are peals? If there were peal towers and quarter towers, this would be bimodal too.

Same shape. Towers differ in how much peal ringing they host, continuously, the way they differ in everything else.

And it is not something you graduate to

A third possibility: one population, but peals are what you progress to, so peal share climbs with experience. It does not.

Mean peal share Share who have ever rung a peal

Flat between 8.5% and 12.1% from one year active to eleven. The twelve-year point is the full width of the corpus, so it means “present throughout” rather than “has rung for twelve years”, and is marked as confounded rather than dropped.

The activity ladder

The share of ringers who have never rung a peal falls steadily as you require more ringing — which is what a single population with a long tail looks like, and not what two separate groups would produce.

Why bad identity resolution invents a split

The same histogram, computed over raw name strings instead of resolved identities. It is the same corpus and the same question; only the notion of “a ringer” changes.

Resolved identities

Raw name strings

Splitting one person across name variants gives each fragment a narrower slice of their ringing, so a fragment can look 100% peal when the person is not. Poor identity resolution manufactures apparent specialists at both ends. The unresolved view is the one that looks more like two populations — and it is the less trustworthy of the two, so better resolution would strengthen this finding rather than undo it.

A correction, left visible

The first draft of the query comment claimed unresolved names could only thicken the left-hand bar. That was wrong, and the pair of charts above is what showed it. The right tail is where the damage is.

What this page cannot see

The corpus is overwhelmingly quarter peals: performances of 1,000–1,399 changes, against peals. The commonest Sunday length is 1,260, tens of thousands of times over.

Ordinary Sunday service ringing is not in this corpus at all. Rounds and call changes for half an hour before a service are not reported to BellBoard; only performances somebody thought worth recording are. So the original question — service band versus peal circuit — asks about a population the data cannot see, and this page answers the version it can: quarter ringers versus peal ringers.

That population, the one that only ever rings for services and never rings a quarter, may well be genuinely separate. Nothing here rules it out. What is ruled out is a split between the two activities the corpus does record.