2026 Ringing Census: implementation design

Version 2.8 · 12 November 2026

Author

Mark Ainsworth, for the census committee

Modified

12 November 2026

Prepared by Mark Ainsworth with Claude (Anthropic). Mark and the committee made the decisions; Claude drafted the text from them, which Mark reviews.

The 2026 Ringing Census will estimate how many active tower bell ringers there are, and who they are, with a range tight enough to measure Ringing 2030 against. It invites every tower and every ringer. Two small random samples, chased hard, then show how the people who replied differ from those who didn’t, so that the difference can be corrected. The design is settled; the details of the sample draw are being finalised this month.

This page sets out the whole plan for readers who were not in the meetings: the analysis volunteers first, then the committee, Central Council delegates, and whoever repeats the census in 2030. Part 1 is the plan on one page. Part 2 tells it step by step, with the reasoning in Why boxes you can open. Part 3 holds the reference material. The colour key is under How to read this page.

Part 1. The census on one page

What we measure, and the problem in the way

The census gives Ringing 2030, the Central Council’s strategy for growing ringing, a trustworthy baseline: the total number of active tower bell ringers, with a stated range, and who they are. The last count that could be trusted was in 1988. The census asks about the reference week, Monday 16 to Sunday 22 November 2026; the surveys open on 23 November.

Answering online costs nothing, so everyone is invited: every tower in Dove’s Guide with four or more bells, and every ringer. But the people who reply unprompted are not typical. Keen ringers, thriving towers and the well connected reply more. Counting only the replies, or scaling them up, gives the wrong total and the wrong make-up of the population. And the replies alone can’t show how wrong they are, because we know nothing about who stayed silent. This is non-response bias.

The fix: chase a random sample, and measure the bias

The census draws a random sample of towers and chases every one until it replies. Enumerators, the volunteers who do the chasing, call these Lucky Dip towers. Because the sample is random, it shows what all towers look like, silent ones included. Comparing it with the towers that replied unprompted shows how the unprompted replies are skewed, and the model corrects every tower that didn’t reply.

The same idea works one level down. In one Lucky Dip tower per enumerator, their Golden Ticket tower, the enumerator helps the captain get every band member to reply. These complete bands show how ringers who reply unprompted differ from those who don’t, on every question at once.

In 1988 the sample replaced the population. In 2026 it calibrates it.

Everyone is invited. Every tower in Dove’s Guide with 4 or more bells (5,743), and every ringer.
Adds: the list the estimates describe.

Voluntary replies. Towers and ringers who reply unprompted.
Adds: most of the data, but tilted towards keen ringers and thriving towers.

Lucky Dip: chased towers. A random sample drawn from all towers, repliers or not, each chased until it replies.
Adds: what silent towers look like, so the tilt in tower replies can be measured.

Golden Ticket: complete bands. One Lucky Dip tower per enumerator, where the captain is helped to get everyone to reply.
Adds: what silent ringers look like, so the tilt in ringer replies can be measured.

Hierarchical model. Learns the tilt from the two samples, and corrects every tower and ringer who didn’t reply.

An estimate with a range. “Between X and Y active ringers”, with a stated level of confidence.

Each layer sits inside the one before it, except that the Lucky Dip is drawn from every tower, not only from those that reply.

Where things stand

The design is settled; the details of the draw are being finalised.

  • Settled: what we count and when (the reference week, 16–22 November); who we ask (every tower in Dove’s Guide with four or more bells, and every ringer); the two surveys; the two random samples that correct for who replies; the fieldwork timetable; and the data protection approach.
  • Being finalised before the draw (by 23 November): the place classification and the method for the draw. How many towers to chase, the reserve, allocation and Golden Tickets were settled on 8 November; how towers are shared out, when the reserve opens and how coordinators are grouped followed by 12 November. All are set out in Lucky Dip, Golden Tickets and chasing.
  • Before the surveys open (23 November): the survey build and its tests, data protection sign-off, the association list, and the short survey links with their holding pages.
  • Before the reference week: the analysis plan, including the precision we can promise, registered in advance.

The whole design at a glance (colours explained in How to read this page):

Part of the design Status In one line
Reference week Locked 16–22 November; surveys open 23 November
What we count Decided Active ringers, four ways, one of them 1988’s
The frame Decided Dove’s Guide, 4+ bells, British Isles and volunteer countries
Tower picker Locked Every ringable ring; built and published
Tower Representative Survey Decided Five-minute Core, chased; Tail optional
Individual Ringer Survey Decided One link for everyone; no names; no health data
Association list Open: minor With the committee for checking
Place types Open Working answer: Draft C
Lucky Dip: how many Decided 1,000, plus two reserve boxes of 500, opened at the cut-off at 40% and 60% spontaneous response
Lucky Dip: how drawn Open Aim settled; method open between two options
Golden Ticket rules Decided One per enumerator, from their own list, chance in proportion to bells
Allocation to enumerators Decided By time offered and nearness to people’s ringing; own-band towers to their member, a week later
Coordinator groups Open Groups of whole associations, about √E; role to settle with Elva
Selection discipline Decided No tower leaves the sample; same message everywhere
The model Decided Hierarchical, giving a range; agreed in outline
Precision and pre-registration Open Before the reference week
Fieldwork timetable and roles Decided Lists out 23 November; chasing from 30 November
Survey links in Census Week Open: minor Holding pages until Monday 23 November; technology open
Enumerators’ details Decided The check-in; only checked-in enumerators get towers
Moving towers when someone drops out Open When is settled; who takes them on is not
Response status for enumerators Decided One email each morning; workbooks for coordinators only
Data protection, young ringers Decided Independent sign-off pending

Part 2. The plan, step by step

Each step builds on the one before. Decided material reads as plain text. Anything still open carries a coloured stripe and a label, with the working answer where there is one.

1. Why now: there has been no usable count since 1988

Ringing has not been counted in a way that gives a defensible total since 1988. Ringing 2030 needs one: the strategy aims to recruit significant numbers of new, younger ringers, and without a baseline nobody can tell whether it is working.

The Central Council has led four attempts in five decades. Only 1988’s gives a total that can be defended.

Year Design What survives Why it does or doesn’t give a count
1972 Thousands of towers; general information about each band Summary statistics on 1970s computer printout, overall and by guild How it was distributed and how replies were gathered is not known
1988 A random sample of 500 towers (about 10%), every one followed up by 77 volunteers; one ringer in three at those towers; separate questionnaires for incumbents, tower captains, ringers and associations A series of Ringing World articles, now a digital edition It gives a count: about 41,000 active ringers, nearer 36,000 once ringers counted at more than one tower are allowed for
2007–8 Thousands of towers; activity in a sample week Tower-level replies in Access and Excel, with some basic summaries No sampling strategy, so the replies cannot be scaled to a total. No record of the planning decisions
2015 A planned 10% sample of ringers from association membership lists None Stopped shortly before launch after objections about data protection

Each attempt teaches something. 1988 worked because it chased a random sample until nearly every tower replied. It used a small sample because every form was on paper, posted and typed in by hand. 2007–8 shows that a pile of replies with no sampling design cannot give a total, however large the pile. 2015 shows that working from lists of named people is a dead end; the 2026 census asks no ringer for their name.

Two things have changed since 1988. First, answering online costs nothing, so everyone can be invited rather than a sample. Second, modern statistics can correct for who responds far better than 1988 could. Election pollsters face the same problem (who answers polls, and who turns out to vote) and use multilevel regression and post-stratification (MRP): a model learns how answers and response differ between groups, and the results are re-weighted to the known make-up of the whole population. The census adapts that approach.

What 1988 did, in its own words

The 1988 committee explained its design in the Ringing World. For towers it chose 500, “picked at random from all of the towers listed in an updated computer version of Dove’s guide, arranged alphabetically.” For ringers, it asked each sampled tower’s captain for a list, “to organise them alphabetically, and then to select every third in the list starting with the second (ie ringers 2,5,8 etc.).”

Its headline: “almost certainly the number of active ringers in the British Isles at the time of the survey was between 38,000 and 44,000, and probably about 41,000.” A later chapter found that up to 20% of the ringers surveyed belonged to joint bands or more than one band, and that the true figure “may be much closer to that figure of one-per-bell (i.e. about 36,000)”.

The full survey is in the digital edition.

2. What we count: active ringers, four ways

The headline is the number of active tower bell ringers. It is counted against three time anchors, because no single one answers every question, and one of them keeps 1988’s definition so the two censuses can be compared.

What the census is for

The primary objective is an estimate of the total number of active tower bell ringers, with a well-characterised range, and the make-up of that population: age, gender, ability, motivation, how and when people learned, and what brings people back. The range has to be tight enough that future recruitment and retention initiatives can be measured against it.

The secondary objectives are:

  • the state of health of towers;
  • the number of active handbell ringers;
  • types of ringing activity, and how often people ring;
  • reasons for lapsing, to support follow-up studies;
  • comparability with 1988 (the roster and Census Sunday series), and calibration against Pealbase, BellBoard, Dove, association membership rolls and the UK census;
  • results returned to the ringing community, nationally and by association, in a form that helps people act on them.

Three time anchors

A fourth anchor, everyone who rang at the tower at least once in the twelve months to 22 November, was dropped on 17 November (D-089): reps found it very hard to answer, and occasional ringers counted at several towers cannot be added up. Towers on a rota, silent in census week, are covered by the band count and the shared-band questions.

Anchor Who is counted Why we need it
Roster Band members on the 1988 definition, whether or not they rang recently The series directly comparable with 1988’s ~41,000 (~36,000 after removing double counting)
Reference week Everyone who took hold of a rope at the tower, for any purpose, 16–22 November The modern, objective measure
Census Sunday Everyone who rang at the tower on 22 November 1988’s landmark findings are all about Sundays

The roster definition is taken almost word for word from 1988, and must not drift: count everyone who rings here on Sundays and/or at practice and whom you consider a member of the band, including irregular and infrequent attenders. Include people away at school or university if they ring here regularly for part of the year. Include learners only if they can ring unaided. Learners who can’t yet ring unaided are counted separately, as the recruitment pipeline.

Two sentences are added after “infrequent attenders” (D-060), without changing 1988’s words: count people by whether they ring with the band, not by whether they are affiliated to this tower in your association’s records or on any church list. If your band also rings at other towers, count the whole band. The first stops the same words being counted against an association’s register or a parish safeguarding list, which some reps now treat as “the band”. The second makes a shared band’s count the same from each of its towers, as 1988’s lists were; the overlap is removed in the analysis from the towers each return names.

The Individual Ringer Survey carries the same definition written from the ringer’s side. It asks whether the tower captain would count you as one of their ringers, not whether you are a “member”, because many regular ringers would not call themselves members (D-062). If either wording changes, both change together.

The reference week: 16–22 November

The reference week is Monday 16 to Sunday 22 November 2026, and Census Sunday is 22 November. The surveys open on Monday 23 November, the day after the week ends, and ask about that week. These dates are locked: they are already in the enumerator materials and the publicity. The survey links themselves go out during the week, leading to holding pages until the surveys open (see Survey links).

Why this week Like 1988’s choice, it is a dull mid-term week. Students are in their term-time towers, and there are no festivals or holidays to distort it.

Open: minor Publicity wording for Census Week.

People may read “Census Week” as the week they fill in the survey, rather than the week they are asked about. The publicity should always pair the two: “Census Week is 16–22 November. The surveys open on Monday 23 November and ask about that week.” The draft Ringing World article of 15 October says the links go out “during Census Week”. Since D-043 they will, but only to holding pages, so the article needs to say plainly that the surveys open on 23 November. It also needs the dropped five-year question removing before it is published. The holding pages can carry the same pairing of dates.

3. Who we ask: every tower and every ringer, through two short surveys

Everyone is invited. Each tower makes one return, from whoever knows its week-to-week picture, and each ringer makes their own. Both surveys run on QuestionPro, and both name towers from the same locked list built from Dove’s Guide.

The frame: every tower in Dove’s Guide with four or more bells

The sampling frame is the list of towers the estimates describe, and the list the random samples are drawn from. It is every ring in Dove’s Guide with 4 or more bells, in the British Isles plus countries where a census volunteer is present. On the 19 September Dove snapshot that is 5,743 rings. Mini-rings are left out of the frame, because most do not host a band that rings nowhere else. The committee agreed the frame on 20 September (D-017, D-019).

The 1988 comparison uses 1988’s own rule: 5 or more bells, British Isles only. On the same snapshot that gives 5,473 rings, within 1% of 1988’s 5,425 towers, which is the evidence that the old rule has been rebuilt correctly.

Outside the British Isles there are volunteers in Australia (5) and North America (3), and none elsewhere. The frame therefore includes Australia and the USA, which add 114 rings.

Returns are matched to the frame on Dove’s ring identifier, RingID (D-007). Dove’s own tower ID is blank for 118 rings, so it cannot do the job.

Open: minor

  • Canada and New Zealand have about ten rings each, in the same associations as the US and Australian towers. Are any of the North American or other volunteers there? (Q-018)
  • Towers with two rings (13 of them): whether the unit is the ring or the tower is deferred (D-010). Two frame towers are affected for bell counts, Gresford and Rugby (Q-020).

Inclusion: anyone can find their tower, wherever it is

Anything that speaks to inclusion covers everyone (D-018, D-040). The tower picker holds every ringable ring in every country, including rings of three and mini-rings: 6,297 rings, among them New Zealand (10), Canada (10) and South Africa (9). A ringer anywhere who looks for their tower and can’t find it reads that as being left out. Estimation needs a frame with a denominator; inclusion does not.

Open: minor The settlement map shows frame countries only (Q-027).

The settlement map is to show every tower too. At present it shows towers in frame countries only.

The tower picker: every tower chosen from a list, never typed

Tower names are vague and often duplicated, so nobody types one. Both surveys name towers through a search box over a locked list built from Dove’s Guide. The picker is built and published. Value capture and iPhone Safari have passed their tests; the remaining tests are under Testing the surveys.

The Tower Representative Survey: a five-minute Core, chased; an optional Tail, never chased

One return per tower comes from the Tower Captain, the Ringing Master, or whoever knows the week-to-week picture (specification v0.4). It comes in two parts, because in 1988 the short ask got answered and the long one didn’t:

  • Core (about 5 minutes, less where there is little ringing): the state of ringing, the roster and its age split, the reference-week, Sunday and twelve-month counts, and learners. This is what enumerators chase to near 100%. It saves as a complete return on its own.
  • Tail (about 5 minutes, optional): bells, practice, reliance on visitors for Sunday ringing, congregation, teaching resources, recruitment and retention. Encouraged, never chased.

The survey’s first question asks who is completing it: the tower’s own representative, or an enumerator on the tower’s behalf. An enumerator skips the privacy acknowledgement and the contact details, and the return is flagged as enumerator-completed. How a return was obtained is itself a model input, because an enumerator’s zero and a captain’s zero are different observations. There are four categories:

  1. completed by the rep, unprompted;
  2. completed by the rep after chasing;
  3. completed by an enumerator with the rep;
  4. completed by an enumerator from observation, where no band is left.

The survey opens with a short privacy notice and a Continue button, not a required tick. The rep’s contact details are optional, with their own tick, and are deleted once results are compiled. Reps correct a return by emailing the central team, not by reopening it.

Why: the 1988 evidence, and why no required tick

1988’s agent-mediated roster count reached 94% response; the full captain questionnaire reached 60%. The short Core is the 94% ask.

The legal basis is legitimate interests (see Protecting people). Under it, the privacy notice must be given, not agreed to. A required tick looks like consent, and it costs returns from anyone who hesitates.

Duplicate tower returns are kept, then one is chosen. They are not blocked on the device, because one enumerator may enter several towers from one computer. Where a ring has more than one return, a full return beats a partial one; otherwise the most recent wins (D-049). The response-status table and the analysis use the same rule.

What we don’t ask: clergy interviews are dropped

1988’s best response rate (93%) came from telephone interviews with clergy, who are also the only authoritative source on congregation health. The 2026 census builds no central contact list (see Protecting people), so there is no route to the clergy of chased towers, and the option was dropped (Mark, 19 October). Congregation health is covered by a Tail question and by sensitivity analysis.

4. The problem: replies lean towards the keen and the thriving

The towers and ringers who reply unprompted are not a fair cross-section, and the replies alone cannot show how far off they are. This is non-response bias, and the rest of the design exists to deal with it.

At tower level, the silent towers are the ones we know least about

Not every tower where ringing happens will reply. Thriving towers and larger bands are more likely to; struggling or silent towers are the least likely to reply of their own accord. That leaves a blind spot exactly where the news may be worst. Count only the towers that reply and the total is wrong. Scale them up and the total is still wrong, because the towers that replied are better off than the ones that didn’t.

At ringer level, the keen and well connected reply first

Not every ringer will reply either. Voluntary respondents will lean towards the keenest and best connected: Ringing World readers, people on association lists and social media. So the replies would overstate how often people ring, how advanced they are and how enthusiastic they feel.

Why ordinary weighting isn’t enough

The usual fix is to re-weight replies to match totals known for the whole population. The census can do that for age, because each captain reports the age split of their band. But age explains only part of who replies. Keenness, ability and frequency matter too, and there is no outside total for any of them. Without some way of seeing the silent towers and ringers, any correction for these rests on assumption.

How we ask also shapes who replies

Who replies depends on how the census is presented as well as on the towers and ringers themselves. Trust in how data is handled, a clear purpose, and a promise of useful results back all raise response. Uneven communication can shift it between groups, which changes the assumptions the model needs. So communication is treated as part of the design, not a detail of delivery:

  • every tower hears the same message, at the same time (see The discipline that keeps the weights valid);
  • the data protection approach is stated plainly, because trust drives participation;
  • results go back to the community, nationally and by association, and associations that share their own tower counts get richer results back.

Strong engagement eases the burden on the model. A technically sound design could still give poor data if trust is low.

5. The fix: two random samples that measure the bias

Alongside the voluntary replies, the census draws two random samples. Each is chased to near-complete response, so each shows what the silent towers or silent ringers look like. Comparing them with the voluntary replies measures the bias, and the model removes it.

Sample Size Drawn from Selection Enumerator task
Chased towers (Lucky Dip) 1,000, plus two reserve boxes of 500 The whole frame Equal chance for every tower; method open Get the Core return, near 100%
Complete bands (Golden Ticket) One per enumerator Each enumerator’s own Lucky Dip towers Chance in proportion to number of bells Help the captain get every band member to reply

Lucky Dip: chase a random sample of towers to near-complete response

A Lucky Dip tower is one drawn at random from the whole frame, before anyone has replied, and chased by an enumerator until it makes a Core return or its status is established. Because every tower had the same chance of being drawn, the Lucky Dip towers are a fair miniature of all towers, silent ones included.

The comparison works like this. Every tower, chased or not, has the week of 23–29 November to reply unprompted, because chasing doesn’t start until 30 November. So within the Lucky Dip we see which towers replied unprompted and which only replied when chased. That gives the voluntary response rate as a function of things we know about every tower: association, place type, ring size and the density of the area. The model applies that relationship to every tower outside the Lucky Dip that didn’t reply.

An illustration, with made-up round numbers. Suppose the frame has 1,000 towers: 500 in small villages and 500 in towns. 600 reply unprompted, 60% overall. The Lucky Dip shows the real pattern: of its village towers, 40% had replied unprompted; of its town towers, 80%. So the 600 replies are two-thirds town towers, although only half of all towers are in towns. Suppose village bands average 6 ringers and town bands 10. Scaling up the raw replies gives about 8,700 ringers. The true total is 8,000. Once each village reply stands for 2.5 village towers and each town reply for 1.25 town towers, the total comes back to 8,000.

The Lucky Dip also shows whether silent towers differ from replying towers of the same type. If silent village towers turn out to have smaller bands, or none, the model corrects for that too. The real numbers will differ; the logic is the same.

Why a fair sample, rather than chasing the towers we know least about

There were two ways to aim the chasing. A representative sample gives every tower the same chance, and asks: how does the chance of replying voluntarily depend on what we know about a tower? The answer can then be applied to every tower that stayed silent.

A targeted sample would over-weight the small, struggling or silent towers, where uncertainty is greatest. Each enumerator hour would buy more information, but the correction would lean harder on modelling assumptions, because the chased towers would no longer be a fair miniature of the whole.

The design takes the representative route. Whether to over-sample small villages and hamlets, which may hold most of the towers without a band, is noted in the sampling chapter. The current plan is not to: it would sharpen that one estimate at the cost of unequal weights.

Why not chase everyone? Enumerator effort is the limit. A volunteer who reaches every one of a few towers does more for the estimate than one spread thinly over many. A chased tower that never replies puts back the very bias the sample exists to remove.

Golden Ticket: complete bands show how replying ringers differ

A Golden Ticket tower is one of an enumerator’s Lucky Dip towers, drawn at random. There, the enumerator helps the captain get the whole band to complete the Individual Ringer Survey. Each complete band is a fair sample of ringers at that kind of tower, keen and reluctant alike.

Comparing complete bands with voluntary respondents from similar towers shows the bias on every question at once: frequency, ability, enthusiasm, and anything else collected. Nobody has to guess in advance which ones matter. This is the logic of Hansen and Hurwitz’s 1946 design for non-response: follow up a random subsample intensively, and let it stand for everyone who didn’t reply. The census adapts it to work through captains rather than lists of people.

An illustration, with made-up numbers. Suppose that among voluntary respondents at six-bell village towers, half say they ring more than once a week. In complete Golden Ticket bands at similar towers, a quarter do. Voluntary replies then over-represent the most frequent ringers about two to one, and the model down-weights them accordingly.

Each enumerator’s Golden Ticket is one of their own Lucky Dip towers, drawn centrally before the surveys open with chance in proportion to the number of bells, the practical stand-in for band size. It is swapped only when its tower return shows there is no band to complete, and then for the next tower in that enumerator’s ordered draw. The full rules, and why they were chosen, are in Lucky Dip, Golden Tickets and chasing.

Why complete bands, and why bells

Why complete bands. Correcting individual non-response by age alone was felt too thin: the keen, frequent, able ringers are the ones who reply, and age captures only part of that.

Why bells. In one association’s 2025 register, registered members rise roughly in line with bells, and towers with few bells more often have no members. Choosing by bells gives every ringer roughly the same chance of being in a complete band, and cuts the number of substitutions. The choice affects efficiency, not validity.

Second-round Golden Tickets are recorded as such, so the analysis can check whether swapping biased anything.

Open: minor Minimum roster for a Golden Ticket (working line: 3).

How the Lucky Dip towers are drawn

Two things are still being finalised before the draw: the place types the draw is balanced on, and the method that balances it. The size was settled on 8 November. The analysis team’s views on the method are in Lucky Dip, Golden Tickets and chasing.

Open Place types. Working answer: Draft C (D-037, Q-023). Needed before the draw.

Each tower gets one of seven place types, built from what is around the church today, never from ring size:

Type How it is found
Small village or hamlet Fewer than 200 people per km² of land within a short walk of the church (a 250 m kernel)
Village 200 per km² or more, but not a town
Town 600 per km² or more, outside a city’s built-up area, with a high street (any mapped retail centre within 600 m of the church) or in a dense town of 5,000 to 50,000 people
City or large town: centre Inside a large built-up area, and within 250 m of a high street (a CDRC District Centre or larger), or with 3+ towers within 1 km
City or large town: suburb Inside a large built-up area, otherwise
Conurbation: centre As for a city centre, in a built-up area of 575,000 people or more
Conurbation: suburb Inside a conurbation, otherwise

A kernel here is a weighted count of the people living around the church, with nearer people counting for more; “250 m” is the distance over which the weight halves. Tower clusters, towers unusually close together (three or more within 500 m of each other, a few minutes’ walk), are flagged separately, because their bands are often shared. Every tower’s type is on the settlement map.

Why this classification, and its weak points

Why not the ONS rural-urban classification, which v2.0 expected to use? It describes areas, not the walk to a church, and it stops at the UK border. The classification here is built from global population grids (GHSL) and the CDRC’s map of retail centres, so it works the same way for every tower and can be compared with 1988’s five types. Shares of comparable towers come out within 3 points of 1988’s (villages 56% against 53%, towns 28% against 28%).

Neutral names. People will look up their own town, so no name grades a place.

Weak points. Churches set back from the high street; the retail map is UK only; the village line is a convenience, not a natural break. Abroad, a centre is found from the share of non-residential buildings instead.

Full reasoning: D-026 to D-037, and the analysis chapter Choosing a classification.

How the draw is balanced (decided, D-074). The Lucky Dip is drawn by the cube method (Deville and Tillé 2004): an algorithm that picks towers at random, each with the same chance, while holding several totals at their fair share at once. With no ranking among them, it balances:

  • every place type x ring-size cell nationally (all seven types, tower clusters as one; ring sizes 4–5, 6–7, 8–9, 10 or more);
  • every association’s total;
  • within each association, its own mix of villages and hamlets, towns, urban centres and urban suburbs, and of rings of 8 or more.

It does not sort or spread, so the draw neither seeks nor avoids tower clusters. Overseas towers are their own stratum, inside the 1,000 core, with twice their proportional share. The core and both reserve boxes are three nested balanced draws, so whichever boxes open, the towers chased are a fair sample. The seed, 5040, was fixed before any draw.

Why this method, and what it cost

Why not fixed quotas. Seven place types crossed with four ring sizes make 32 cells, many of which would expect very few chased towers and need arbitrary merging rules. Balancing holds the cells at their fair share without needing a minimum in each.

Why balance within associations. The figures people will look at most are their own association’s. A draw that only fixes an association’s total can, by chance, give a small association all villages and no towns. Balancing its own mix of places and rings means its figures stand on a fair mix of its own towers.

How it was chosen. Eight methods were compared by simulation (analysis/sampling-design.qmd), on two stand-in populations from 1988 and one shaped by an association register. This method was the best of all for association figures and for balance within associations, best or near best on every national fair share, and within noise of the best nationally. The alternative logged earlier (D-042), a systematic draw from a list sorted by association, place type, ring size and density, was more precise nationally on the register-shaped stand-in, but missed ring-size shares by about 9 towers, ranked the three variables, and kept neighbouring towers apart.

Standing. The cube method is the general form of stratified sampling. INSEE, the French statistics office, draws its census rotation groups with it.

Variance. It has no simple textbook formula for its sampling error, and needs an approximation. The estimates are model-based, so this matters little.

The job of the sample is calibration, not replacement: in 1988 the sample replaced the population; in 2026 it calibrates it.

Size: 1,000 towers, with two reserve boxes of 500 drawn at the same time and opened, whole, at the 9am cut-off on 30 November only if many towers have replied on their own. The reasoning, including why spare effort goes into completeness rather than more towers, is in Lucky Dip, Golden Tickets and chasing.

Allocation: location decides who chases a tower, never which towers are chased

The analysis team shares the drawn towers among enumerators who have checked in. Lists follow the time each person offered, within limits, and towers go as near as possible to where people ring: their home, their home tower, or any other tower whose band they belong to. Effort rises with distance the way relationships fade, from 1 within 5 km to 3 at 20 km, and stays nearly flat beyond that within the enumerator’s own association. Distances go round the Severn, Humber and Solway. A tower whose band an enumerator belongs to goes to them, but is released on 29 November, so the voluntary week stays clean. Location can shape who gets which tower, never which towers are drawn. The rules and the evidence for them are in Lucky Dip, Golden Tickets and chasing.

Enumerators choose themselves, by volunteering and then confirming. That does not bias the design, because towers are drawn from the whole frame before they are allocated.

Open Moving towers when an enumerator drops out. When a tower moves is settled; who takes it on is not. Both are set out in Lucky Dip, Golden Tickets and chasing.

The discipline that keeps the weights valid

The samples only correct the bias if nobody bends them. A design weight is the number of towers, or ringers, that each sampled one stands for. A chased tower’s weight is 1 / P(chased). A ringer in a Golden Ticket band carries 1 / (P(chased) × P(Golden Ticket)). Those weights mean something only if the following rules hold:

  • The selection code is committed to the public repository before the draw, with the random seed recorded.
  • Every tower’s selection probabilities are stored alongside the allocation.
  • No tower ever leaves the sample, by anyone, for any reason. Lucky Dip towers stand in for the towers nobody phones, so an awkward one can’t be dropped or traded for an easier one. Towers move between enumerators only through the central team.
  • A Golden Ticket is swapped only when it has no band, and only for the next tower in that enumerator’s ordered draw.
  • No warmer contact before chasing starts. Until 30 November, chased towers hear exactly what every other tower hears. Tower lists are drawn centrally beforehand and released only on 23 November, so enumerators can’t treat their own towers differently even by accident.
  • Standard reminder only at Lucky Dip towers other than the Golden Ticket. Enthusiastic people will want to push everywhere. If they do, the comparison that makes the Golden Tickets useful disappears.
Why the same message everywhere

Everyone knowing the census is coming (uniform reactivity) raises every tower’s count by roughly the same amount, and the model can live with that. Chased towers getting warmer, earlier contact than the rest (differential reactivity) can move the very thing being measured, not just who replies. That is the distinction the whole timetable is built to preserve: awareness before the week, engagement only after it.

Extra ringing for the census would inflate the reference-week and Sunday counts, even if it happened evenly. That is why, during the week, “encourage participation” means notice and note who rings, not come and ring so you’re counted. The roster count is immune; the 1988 Sunday comparison is not.

Spreading the word will be stronger where enumerators live. That has nothing to do with which towers are chased, but it can lift voluntary response locally, so enumerator density by area is recorded as a model input.

6. From data to estimate: a model that gives a range

A hierarchical model brings the layers together and gives a range, not a single number: “there are between X and Y active ringers”, with a stated level of confidence. The design of the model is agreed in outline; its precision and the analysis plan are still to fix.

The model mirrors how ringing is organised

The model is hierarchical because the data is: ringers sit in bands, bands in towers, towers in branches and associations, and those in regions. The analysis team builds it, starting from a generative model, a simulated version of the census that lets the design be tested before any data arrives. Regional, association and place-type effects stay in as partially pooled terms: each group’s estimate borrows strength from the others, and an effect shrinks towards zero if the data doesn’t support it.

At tower level, the Lucky Dip gives a near-complete picture of a random slice of the frame, including towers that would never have replied on their own. Comparing it with the voluntary replies gives the chance of replying as a function of tower characteristics (association, place type, ring size, density). That relationship is applied to every tower outside the Lucky Dip that didn’t reply.

At ringer level, each ringer’s return is tied to its tower through the tower list, and the captain’s roster gives the denominator. Three lines of evidence check one another:

  1. Age. The captain’s roster gives the size of each age band, so ringer returns can be weighted to it. Necessary, but it only corrects for what age explains.
  2. Complete bands. Golden Ticket towers show the whole band. Comparing them with voluntary respondents in similar towers measures the bias directly, on every variable.
  3. Timing. Early responders tend to be keener than late ones; the Ringing Trends Committee’s survey with the Yorkshire Association showed exactly this. The complete bands test whether extrapolating that timing gradient holds.

Measurement checks

  • Captain counts are cross-checked against individual returns. 1988 found that turnout averaged about 55% of band size.
  • Ringers counted at several towers are resolved by the home-tower question and the shared-band question. 1988’s allowance for this took its headline from ~41,000 to ~36,000.
  • Congregation health is the main confounder the census can’t observe directly. If the Tail question proves unusable, it is handled with sensitivity analysis rather than ignored.

Open: minor Outside benchmarks.

  • Association membership data is assumed usually missing. It is sought mainly after the census, as a carrot: associations that share tower counts get richer results back. Where it exists, the Lucky Dip towers give the ratio of real roster to registered members, partially pooled across associations, because those that share are self-selected.
  • Known keen subgroups. Pealbase gives the number of distinct people who rang a peal in the last 12 months; BellBoard gives about 4,600 distinct people a year in Surprise and Stedman performances, and about 3,300 a year meeting the Red zone rule (two or more such performances, at least one on an inside bell; D-045). The second is the benchmark for ringers who say they rang one of the listed methods in the last 12 months. The share of each who reply shows how steep the keenness gradient is. Both are used at aggregate level only, never for matching individuals.

Open Precision target and pre-registration. Needed before the reference week.

  • Precision target. v1 proposed a 90% interval of about ±15% on the national total. Simulations from the generative model should confirm whether that is achievable.
  • Pre-registration. The analysis plan is to be registered before the reference week, so nobody can say the weighting was tuned after seeing the results.

7. How it runs on the ground: awareness before the week, chasing after it

Before and during the reference week, every tower hears the same thing. Enumerators get their tower lists only once the week is over, and start chasing a week after that. Everything about the fieldwork is designed for volunteers and ringers who are not confident with technology.

One timeline

The surveys open on Monday 23 November. Enumerators get their tower lists that day and start chasing on Monday 30 November. The stated deadline is Sunday 6 December; chasing and late returns carry on to 21 December.

The gate: no tower is contacted until data protection sign-off is complete and the QuestionPro build has been tested end to end.

Phase Central team Enumerators Tower reps Ringers
To 25 October Uniform awareness One notice through The Ringing World, association newsletters, CCCBR channels and the website: same content, every tower. Collect role confirmations and enumerators’ details Confirm their role. Spread the word, the same way everywhere Receive the notice. Asked to note who rings in Census Week, especially on the Sunday Same notice
26 October to 15 November Dry run and feedback Support the dry run. Reallocate the lists of anyone who steps down. Fold in survey feedback Dry run on the test survey (to 8 November). Keep spreading the word No change Feedback waves on the survey
16–22 November Census Week Publish both short survey links and their QR codes, leading to holding pages. No other outbound contact. Check-ins close (19 November); draw the core, reserve boxes and Golden Tickets Encourage noting who rings, not ringing more Band rings as normal and notes attendance Ring as normal. The survey link shows a holding page until the surveys open
23–29 November Surveys open; enumerators prepare The short links switch to the live surveys, on a schedule set in advance. Email each enumerator their towers Still no contact. Learn their towers; work out how to reach each rep Complete the Core and their own individual return Complete the individual survey, even if they no longer ring
30 November to 6 December Chasing 9am cut-off on the 30th; open reserve box A if spontaneous response has reached 40%, B as well at 60%. Own-band towers appear. One email to each enumerator each morning Follow up chased towers with no Core. Help the Golden Ticket band reply. Standard reminder elsewhere Chased reps followed up. Golden Ticket captains see running totals Reminders through general channels
8–21 December After the deadline Late returns accepted and flagged Final push, finishing on 21 December so the last Sunday before Christmas can be used Late returns accepted Late returns accepted
From 22 December Close and analyse Coverage review, thank-yous, data cleaning. Association membership requests in January Feedback on recurring issues Results fed back Results fed back

The week of 23–29 November does two jobs. It is a clean week of voluntary response from every tower, before any chasing, which is what the Lucky Dip comparison needs. And it is the enumerators’ preparation week.

Two volunteer roles

Role What it involves Tower list
Census Ambassador Spreads the word before and during Census Week, and answers questions locally. Needs no technology None
Enumerator Everything an Ambassador does, then chases their Lucky Dip towers and helps their Golden Ticket band from 30 November A list sized to the time they offered (1,000 towers among everyone who checks in: four or five on average), released 23 November; own-band towers on 29 November

Taking on the Enumerator role is a self-selection, in two steps:

  1. A four-point confirmation (about 20 October). Enumerators confirm they can read email and reply within a couple of days; open an emailed list and a web link; fill in a short online form while on the phone to someone; and give two to three hours a week from 23 November to 21 December. Anyone who can’t tick all four is pointed to the Ambassador role.
  2. A dry run (26 October to 8 November). Each enumerator completes a practice tower return for their own tower on a test copy of the tower survey. Anyone who struggles gets help, takes a buddy, or becomes an Ambassador without fuss.

The check-in. Enumerators give us what allocation needs through a short QuestionPro check-in: where they are, their home tower, their availability through fieldwork, and separate permissions for sharing their details. Only those who have checked in, by 19 November, are given towers.

What enumerators do

Where What the enumerator does When
Everyone in their area (Ambassadors and Enumerators) Make sure all local ringers and towers are expecting the census, using the standard materials. The same message everywhere Before Census Week
Everyone in their area Encourage participation and answer questions, the same way everywhere. Participation means noting who rings, not ringing more than usual Census Week, 16–22 November
Their Lucky Dip towers Learn which towers they have. Work out how to reach each rep, from published contacts or people they know. No contact yet 23–29 November
Lucky Dip towers with no Core return Follow up the rep. Help in practical ways, including completing the Core with reps who have no computer. Where no return can be made, establish and record the tower’s status. Goal: every allocated tower accounted for From 30 November
Their Golden Ticket tower Give the captain the tower’s survey link and the running total of replies, and work with them to get everyone in the band to reply. Goal: every ringer in the band accounted for From 30 November
Their other Lucky Dip towers Pass on the standard reminder about the individual survey, and nothing more From 30 November

The last row is the one the design depends on: without it, the comparison that makes the Golden Tickets useful disappears.

There is no central contact list. Enumerators find their own route to each tower’s rep, using only published association contacts or people they already know, and they do not build or keep lists. Every first contact uses a standard message that links to the privacy notice. The reason is in Protecting people.

Rules for the enumerator deck and handbook:

  • No tower leaves the list. An awkward tower can’t be dropped or traded for an easier one. Towers move between enumerators only through the central team.
  • Standard reminder only at every Lucky Dip tower except the Golden Ticket.
  • Census Week wording: notice and note who rings, not ring more than usual.
  • Totals, never names: the Golden Ticket captain is told how many have replied, never who.
  • Safeguarding: chase through the tower rep only; never chase individual ringers, and never contact anyone under 18.

What captains see: totals, never names

  • At a Golden Ticket tower, the captain sees a running total only: “5 of your 11 band members have responded”. Never an age split, never names, never anything about under-18s.
  • Encouragement is persuasion, not pressure. The individual survey is voluntary, and a polite no is a valid outcome.

What enumerators see: one email each morning

Each enumerator gets one email each morning when there is something to do, and every Monday: their towers still without a return, what to do about each today, their Golden Ticket’s running total, and who to call for help. SharePoint workbooks proved too hard for many enumerators to open (committee, 7 November), so they are kept for coordinators and the central team only. The rules for the emails are in Lucky Dip, Golden Tickets and chasing, and every word of them is in config/email_text.md.

Two rules from the earlier design carry over. No return status is shown for towers that are not chased, so voluntary returns stay voluntary. And ringer counts are shown only for Golden Tickets (D-050); a ringer counts at every tower they name, so a shared band can pass 100%, and the figure guides chasing rather than measuring anything (D-049).

Designed for low technical confidence

Most enumerators and most ringers are assumed not to be confident with technology. Any step that needs a browser skill, an online account or a shared spreadsheet needs a plain alternative.

  • Phone and paper are first-class routes. Paper carrying personal details is kept safely and shredded once entered.
  • One link per job. Enumerators get a single link for tower returns.
  • Simple materials: a one-page card per task in large type, a short recorded walkthrough, and a named person to phone.
  • Accessible surveys: QuestionPro’s accessible theme, large text and short pages.
  • Dropouts expected. An absent enumerator’s open towers are moved on through the central team. No tower quietly falls off the list.
  • Reach beyond older networks: university societies, social media, youth groups, and captains and parents for the youngest.

Open: minor The printable version of the individual survey.

Testing the surveys: in waves, ending with the enumerators

Feedback comes in waves, as agreed by the committee on 16 October. Elva’s first respondents have already commented. Each committee member is gathering more from their own networks before the next meeting. The analysis volunteers test the tower picker. Enumerators are the final wave, through the dry run.

Open: minor Tower picker improvements to test (Q-002, Q-011, Q-013).

Value capture and iPhone Safari have passed. Still to do: going back a page (does the picker reload?); checking the several-towers picker’s values in an export; the several-towers picker’s layout on desktop and tablet; what it shows while the list loads; reducing effort when many towers are named; and large system text sizes and older phones.

8. Protecting people: no names from ringers, no special category data, no contact with children

The census asks no ringer for their name, collects no special category data, and never contacts anyone under 18. The committee accepted the data protection review’s recommendations on 16 October. Independent sign-off is still pending, and nothing here is legal advice.

Data protection

  • Legal basis: legitimate interests (UK GDPR Article 6(1)(f)), with a documented balancing test. Public task is closed: the Data (Use and Access) Act 2025 requires it to be laid down in law.
  • No special category data is asked for: no religion, ethnicity or health. Tower and church information is not used to infer religious belief. Anything volunteered in free text is redacted before analysis, and free text from under-18s is deleted.
  • Individual ringers give no name or contact details. Tower reps’ contact details are optional, on the tower return only, and deleted once results are compiled.
  • No central contact list. Enumerators use only published contacts or people they know, and keep no lists.
  • The response-status table holds tower-level facts only. The daily pull from QuestionPro is reduced to towers in memory and is not saved; any temporary file is deleted in the same run (D-048). Only named snapshots, such as the final export, are kept.
  • Storage: QuestionPro’s EU data centre, so “held in the UK or EU”.
  • Right to object is stated clearly and separately in the opening notice.
  • Enumerators and analysts sign a confidentiality undertaking.
  • Aggregate results are kept indefinitely; published figures never allow re-identification. Individual returns are treated as identifiable throughout, because home tower plus age band plus experience can pick someone out.
  • Captain counts: a captain need not drop a band member from aggregate counts on request, but the band must be able to see what is being done and why.
  • Controller: the CCCBR.
  • International: overseas collaborators work only with pseudonymised extracts.
Why there is no central contact list v2.0 had the central team filling contact gaps for every tower by phone. A central list would mean collecting thousands of people’s details from third parties, and telling each of them where we got their details from (UK GDPR Article 14). Enumerators using published contacts or people they know avoids that, and the standard first message carries the privacy notice.

Open: minor

  • Risk scores in the DPIA: Vicki is circulating the agreed version.
  • Independent sign-off (Q-009).
  • Under-18 age cut-off for participation through a parent, guardian or captain (13 or 16).
  • Whether published outputs are “adapted material” under Dove’s CC BY-SA licence (Q-008).
  • Click logging by the short-link service, if any (Q-029).
  • Showing tower status to enumerators, and the pull’s temporary cache, to be covered in the DPIA (Q-016).

Young ringers: counted and surveyed, under tighter rules

Young ringers are central to Ringing 2030, so they are counted and surveyed, under tighter rules than adults. This is decided in principle.

  • Counting. The captain’s headcount gives under-18s as a number only. No names, ever.
  • Surveying. Age band, never date of birth. No name, contact details, address, school or photograph. Free text from under-18s is deleted before analysis. A privacy notice written for a child to understand.
  • Contact. No enumerator or project representative ever contacts a ringer under 18.
  • Outputs. Higher minimum cell sizes for any published figure involving under-18s.
  • Early starters. The survey records age started, current age and current status, so a real effect of starting young can be told apart from survivorship.

Part 3. Reference

Known complications

Complication Handling Status
Tower identity. Tower names are vague and duplicated Locked list of towers from Dove, chosen through a search box, never free text Locked built and published; tests in Testing the surveys
Rota and benefice bands. One band of 13 ringing six towers on a bi-monthly rota may show one practice of five in the week Roster and twelve-month anchors are immune to the rota; the shared-band question records that the band rings elsewhere. How such a band reports needs a written rule Open: minor
Cyclic towers ringing every four or six weeks A reference-week zero is a real answer, not a non-response. The twelve-month anchor carries the signal Decided
Itinerant ringers Home tower on the individual survey, plus de-duplication at analysis. Edge cases need a defined rule Open: minor rule to write
Society bands, such as the College Youths’ London towers The society’s steward completes the return for that tower Decided; who reports where no steward exists is Open: minor
University ringers Mid-term reference week; students count at their term-time tower and, on the roster definition, at home if they ring there regularly for part of the year Decided
Captain optimism Cross-checked against individual returns and 1988’s 55% turnout Decided
Chiming, carillons and clocks. Some towers have their bells chimed (often on an Ellacombe apparatus) where full-circle ringing is rare or absent Out of scope: the census counts full-circle ringing only (D-066). Survey wording says so at every count, including the Sunday question for towers with no band Decided

Open decisions, in the order they are needed

Owners are suggestions until agreed. Items carried over from 4 October whose progress has not been recorded here are marked (carried over).

Decision Status Owner Needed by
Place classification (Draft C) confirmed for the draw Open Analysis team Before the draw
Balancing method for the draw Decided Mark (D-074, 13 November): cube method, balanced nationally and within associations Before the draw
Minimum roster for a Golden Ticket (working line: 3) Open: minor Analysis team Before the draw
Canada and New Zealand volunteers (frame countries) Decided Mark (D-075, 12 November): both in the frame Before the draw
Two-ring towers: ring or tower as the unit Open: minor Committee Before analysis
Method follow-up questions: wording agreed with Simon Linford and Elva Open: minor Mark, committee Before the build is frozen
Association drop-down: 86 options, the Council’s societies plus university societies (D-059, D-064); Guild of Clerical Ringers listed with no inference about religion (DPIA); Friends of Dorothy not listed, as not a Council member Decided Committee, Mark Done 31 October
Reporting rule for rota and benefice bands Open: minor Mark, Elva Before the build is frozen
Paper version of the individual survey, and rules for paper returns Open: minor Elva, data protection workstream Before the surveys open
Accessible theme and large text switched on in both surveys Open: minor Vicki Before the surveys open
Live-build tests: several towers from one computer; paper returns; Golden Ticket link; cookie blocking Open: minor Vicki Before the surveys open
Tower picker tests and improvements Open: minor Mark, analysis volunteers Before the surveys open
Data protection sign-off, DPIA risk scores, under-18 cut-off Open Data protection professional, Vicki Before the surveys open
Who reports for society-band towers with no steward Open: minor Committee Before the surveys open
Ringing World article: pair “Census Week” with the opening date; remove the five-year question Open: minor Vicki Before it is published
Short links, redirector, holding pages and QR codes (Q-029) Open: minor Committee, CCCBR technical team Before printed materials go out
Pre-register the analysis plan Open Analysis team Before 16 November
Precision target confirmed by simulation Open Analysis team Before the reference week, ideally
Who takes on towers when an enumerator drops out Open Mark, analysis team Before 30 November
Allocation settings (effort curve, baseline, hours cap, experience factor, slack) and the overseas draw weight reviewed (D-071) Open: minor Analysis team Before 23 November
Naming enumerators in neighbouring associations in the first email Open: minor Committee 13 November
What coordinators look after, and the groups (D-073) Open Mark, Elva Before 23 November
Morning emails: test with other mail providers; connect the live QuestionPro status; the enumerator survey link Open: minor Mark Before 23 November
Stated deadline, association “carrot” line, distribution plan, channels for young ringers, standard materials, task cards (carried over) Open: minor Vicki, Gemma, Mark Early to mid October
Pealbase and BellBoard keen-subgroup benchmarks; a reproducible BellBoard fetch that meets data protection rules (Q-030) Open: minor Analysis team Before figures are published
Map shows towers in every country (Q-027) Open: minor Mark Before the map is next shared widely

Team

  • Committee: Elva (instrument design), Vicki (QuestionPro build), Tina (CCCBR President), Mark (methodology and planning lead). Gemma supports communications.
  • Analysis team: Mark Ainsworth (lead), Richard Heys, Matthew Greenaway and Ed Gosden (methodology advice), and the analysis volunteers who came forward with the enumerators.

Version history

Version Date Author Changes
2.8 12 November 2026 Mark Ainsworth Allocation brought in line with the companion page’s new section 5: lists follow time offered (reversing the 7 November set-aside for the first allocation), effort follows distance from people’s ringing, with estuary detours (D-071); own-band towers go to their member, released 29 November (D-070, replacing D-069); reserve boxes open at 40% and 60% spontaneous response, shared on top of existing lists (D-072); coordinators look after groups of whole associations, about √E of them (D-073, open with Elva). Golden Ticket chances now vary with list length and are recorded for weighting
2.7 8 November 2026 Mark Ainsworth Aligned with the new companion page Lucky Dip, Golden Tickets and chasing, which now holds the detail on sampling and chasing. Lucky Dip of 1,000 with two reserve boxes of 500; only checked-in enumerators get towers; Golden Tickets from each enumerator’s own list, with no separate reserves; one email each morning replaces enumerator workbooks, which are kept for coordinators. The Core is described as about 5 minutes. Published deadline 6 December. Reassignment after dropouts is open
2.6 31 October 2026 Mark Ainsworth What enumerators see is now one read-only workbook per association area, with a prototype for feedback: five sheets, a map of each area and its neighbours, updates every two hours (hourly at busy times), towers shown only once released. Open points added for publishing to SharePoint, the central control file and the enumerator check-in. Working assumption of 1,000 chased towers (D-065)
2.5 31 October 2026 Mark Ainsworth The association list is settled: the Council’s member societies (committee, 30 October, D-059) and university societies from universityringing.org (D-064). Friends of Dorothy no longer listed, as not a Council member. New analysis chapter on the association list and on how each tower gets one association (D-063)
2.4 28 October 2026 Mark Ainsworth Proposed follow-up questions on which methods people can ring, with the BellBoard evidence behind them. Red zone benchmark added to the measurement checks (D-045). Open decisions added for the method wording and the BellBoard fetch (Q-030). The Church of England has agreed that maps from its diocese boundaries, and the derived territory file, may be published with its credit (D-051). Analysis chapters on ringing zones and association territories listed. No earlier decision changed
2.3 28 October 2026 Mark Ainsworth New section on what enumerators see: a read-only response-status table in the CCCBR’s OneDrive, with ringer counts only at Golden Tickets and activated reserves (D-047, D-050). Rule for duplicate tower returns settled (D-049). Golden Ticket assignments, activations and swaps logged with dates. Daily pulls are not kept (D-048)
2.2 20 October 2026 Mark Ainsworth Rewritten as a story around the central idea, with a diagram of the layers and one status table; colour kept for open points only. The balancing method for the draw is shown as open between a sorted systematic draw and a balanced sample. The association list is with the committee. Survey links go out during Census Week, leading to holding pages until the surveys open on Monday 23 November (D-043, D-044). The layers diagram is now styled Markdown. No earlier decision changed
2.1 19 October 2026 Mark Ainsworth Moved from Google Docs to this page, with status colours. Brings in the committee meeting of 16 October (Elva, Vicki, Tina, Mark) and decisions since: sample size, sampling method, Golden Ticket rules, data protection, survey changes, no central contact list, incumbent interview dropped
2.0 4 October 2026 Mark Ainsworth Branched from v1 and restructured into one design, from decisions through working assumptions to implementation
1.x 24 March to 4 August 2026 Mark Ainsworth, Vicki Chapman v1 Methodology Outline: objectives, rationale, outline design, gotchas, data protection, pilot objectives, first timeline

The v1 Methodology Outline stays frozen as the record of why the design first looked the way it does. This page records what we are actually doing.

How to read this page

The table in Part 1 gives every major part of the design one of four labels. In Part 2, only open points carry a label, with a stripe of the same colour down the left of the text it covers; everything without a stripe is decided.

Label Meaning
Locked Decided, and already built, published or announced. Changing it now would cost real work or credibility
Decided Agreed by the committee, or settled in an instrument specification. Could still be revisited
Open A substantive point not yet settled. Where there is a working answer, it is given, marked as such
Open: minor A smaller point not yet settled, or a detail inside a decided point

Decision numbers such as D-038 point to the project’s decision log, which keeps the full reasoning for each analytical decision and is never edited after the fact. The decision register shows the current state of each one. Question numbers such as Q-026 are open questions in the same register.

Where this page and an instrument specification disagree on wording, the specification wins. Where they disagree on design, this page wins and the specification needs updating.