Change Ringing Corpus · Roadmap item 22 · A negative result
performance_ringers.bell records which bell each person
stood behind. It is populated on rows and, until this
analysis, nothing in this project had ever read it. It is the most underused column
in the corpus, and it is the only one that can watch an individual ringer’s arc.
Every ringer knows the supposed shape of that arc: you learn on the treble, you move to the inside bells, and the tenor — and the conducting — come later. It is the kind of thing everybody says and nobody has checked at scale.
Bell number alone will not do it. The tenor of a six is the 6 and the
tenor of a twelve is the 12, so every appearance is normalised by the number of bells
rung in that performance: position = bell ÷ ring size. The treble is
near 0, the tenor is exactly 1. Handbell pairs ('1-2') are a different
activity and are excluded throughout.
If ringers worked up the ring, the difference between where a ringer rings late in their career and where they rang early would be positive — the whole distribution shifted right of zero. It is not. It is centred on zero and very nearly symmetric.
So the strong folk model fails. The obvious weaker one — that ringers find a bell and stay on it — fails too, and in the opposite direction.
Over a career the median ringer covers of the ring, and only spend their last stretch on a single bell. Ringers do rotate through the bells, which is the grain of truth in the folk model — but the rotation has no direction. They move around without moving up.
The other half of the folk model is that conducting comes late. Whether that is true depends entirely on what you mean by conducting, and the distinction is large enough that getting it wrong changes the answer by a factor of three.
Conducting something is ordinary and happens early — most who ever conduct have done it within their first dozen appearances. Conducting a peal is a different activity: a fifth of the cohort, reached three times further into a career.
This page’s source document got that wrong first time. It reported
the “any performance” row and described it as peals — 72.5% conducting a peal with a
median wait of 11 appearances, when the true peal figures are 19.8% and 37. The code
never mentioned peals; the prose written over it did. It is corrected here and in
docs/ringing_careers.md, which says so itself.
A ringer’s last appearance is in the data. An absence is an absence — not a death, not a resignation. The corpus cannot tell someone who stopped ringing from someone who moved, changed their name, or rings at a tower that does not report to BellBoard. These are cohort rates and nothing else; no individual is named anywhere on this page.
Roughly half of everyone who appears and is recorded here is not recorded five to eight years later. Committed ringers attrite at under a tenth of that. The 2020 line crosses COVID, when ringing stopped almost entirely for a year, so these cohorts are partly a pandemic effect rather than a steady rate — and cohorts first seen after 2020 cannot by construction have left before it, so they are omitted rather than shown as zero.
Ordinary service ringing, if it were ever recorded. Exactly as in the two populations result, the ringing the folk model is really about — the Sunday band working a learner up the ring — is not in this corpus at all. This rules the progression out for reported ringing, which is not the same as ruling it out for ringing.
Better identity resolution. The canonical identities are a candidate dataset whose accuracy is unmeasured. Splitting one ringer into variants shortens each fragment’s apparent career and narrows its slice of the ring — which would manufacture both more specialists and more early leavers. The raw-name cohort is fatter than the canonical one, exactly as fragmentation predicts.
A longer corpus. Thirteen years is enough to see no progression. It is not enough to see a lifetime.