Where Attendance Actually Collapses, and What to Design Around It

Ask anyone who has run a weekly group for isolated people where it falls apart, and you get the same answer before you finish the question: not week one, not the last week — somewhere in the middle of the run, after the novelty has worn off and before the habit has set. Practitioners call it different things. The pattern is the same everywhere it’s described: a sharp drop, not a slope.

This matters for a specific reason. Most programme evaluation measures outcomes at the end, or averages attendance across the whole run. That tells you whether people who stayed benefited. It tells you almost nothing about the point where the people who didn’t stay decided to leave — which is the point you’d actually need to redesign around.

The published evidence on this is thinner than the sector’s confidence about it. But three different classes of organisation have, in effect, built their model around avoiding this point rather than fixing it after the fact. Looking at what they each did is more useful than another general essay on dropout.

The clinical trial response: shrink the dose, fix the structure

The intervention literature on loneliness is dominated by small, uncontrolled programme evaluations. The randomised controlled trials that exist are worth reading closely precisely because they had to solve the retention problem to get a clean result — a trial with heavy attrition doesn’t produce usable data.

The HEAL-HOA trials took two different approaches to this and it’s instructive that they didn’t converge on the same one. The 2024 trial tested prosocial engagement and volunteering against a control among lonely older adults in Hong Kong. The 2026 trial, run among 1,151 older adults living in poverty, alone, and digitally excluded, used something much narrower: eight 30-minute telephone sessions of behavioural activation and mindfulness, delivered over a single month, by trained laypeople who were themselves older adults with lived experience of loneliness. That second design reduced loneliness at 12 months by more than a comparison arm using traditional befriending — and it did so with a fixed, short, front-loaded dose rather than an open-ended relationship.

A separate randomised trial of befriending in residential aged care found real effects too — a 2.39-point reduction on the UCLA Loneliness Scale at 8 weeks, 2.71 points at 16 weeks — but it was outperformed head-to-head by the structured approach in the 2026 trial. Read the two together and the finding is not that befriending doesn’t work. It’s that an open-ended commitment is harder to sustain than a bounded one, and the bounded one won when tested directly against it.

The design lesson isn’t “run shorter programmes.” It’s that a fixed, known endpoint changes what a participant is agreeing to. Eight sessions in a month is a commitment a person can hold in their head. An indefinite befriending relationship is not — and indefinite commitments are exactly what collapse at the point where motivation dips and structure hasn’t yet replaced it.

The social prescribing response: struggle with the same problem, less successfully

Social prescribing link-worker programmes face this collapse point constantly, and the evidence base is honest about not having solved it. Systematic reviews report real benefits where people stay engaged — gains in self-esteem and confidence, and a meta-synthesis of qualitative evidence found that what participants valued went beyond contact itself to restored participation and purpose, with structured, purposeful activity outperforming unstructured contact. But a 2025 review protocol notes plainly that despite growing adoption, the effectiveness of social prescribing for older adults remains unclear, and only one peer-reviewed randomised trial exists in the space. The programmes that get referred to are heterogeneous by design — a walking group, an art class, a men’s shed — and that heterogeneity means no one has isolated what happens at the session-by-session level across the sector. Link workers know the drop-off point from experience. The evidence base hasn’t caught up to giving them a fix.

The consumer platform response: remove the thing people drop out of

Meetup and Timeleft solve the problem differently, by not asking for continuous enrolment at all. Meetup’s model, running since 2002 with a reported 60 million members, is built on recurring events a person opts into individually — there’s no programme to fall out of, only a next event to skip or attend. Timeleft runs weekly dinners matching strangers by personality, with no obligation beyond the single dinner someone has claimed a seat at. Both platforms report large scale — Timeleft says it has seated more than 3 million guests — but scale here says more about low friction than about depth of connection sustained over time. Removing the collapse point structurally is not the same as building durable relationships through it.

What the evidence actually supports

Claim	Evidence status
Attendance drops sharply at a mid-programme point rather than declining evenly	Widely reported by practitioners; not directly quantified in the trial literature reviewed here
A short, fixed-dose structured intervention outperforms open-ended befriending	Supported — head-to-head RCT comparison in the 2026 HEAL-HOA trial
Structured, purposeful group activity produces more benefit than contact alone	Supported by qualitative meta-synthesis; not yet by RCT evidence
Social prescribing reliably reduces loneliness across programme types	Weak — reviews report positive outcomes but note limited trial evidence and high heterogeneity
Drop-in, low-commitment formats sustain deeper connection over time	Unsupported — scale data exists, retention-to-depth data does not

What this means in practice: if you’re designing a group programme, decide before it starts how long the commitment is and say that number out loud to participants in week one. An unstated open-ended run gives people nothing to hold onto at the point their motivation dips. A named, bounded dose — eight sessions, six weeks, one term — gives them something to finish. If you can’t commit to a fixed length because the group is meant to be ongoing, consider whether it should instead be structured as a series of standalone sessions someone opts into individually, closer to the Meetup or Timeleft model, rather than a programme with an implied continuous membership nobody agreed to.

What this does not solve

None of this addresses reach. Trial participants, social prescribing referrals, and platform users are all people who had already engaged with something — a clinician, a link worker, an app. The collapse point discussed here happens after someone has already shown up once. It says nothing about the much larger group who never show up at all, and fixing retention among the willing does not touch that problem. The trial evidence is also site-specific: older adults in poverty in one city, older adults in residential aged care in another. Whether a fixed eight-session structure works as well for a workplace peer group or a youth programme is an open question, not a settled one.