Why Drop-In Social Groups Look Successful Right Up Until They Collapse

A weekly drop-in group for isolated or lonely people almost always follows the same arc. Session one: twelve people, high energy, a facilitator introducing everyone. Session two: nine or ten, still promising. Session three: eight. By session four or five, the group has quietly become five or six regulars, and it will stay roughly that size — or shrink further — for the rest of its run. Programme managers see this pattern so often it barely registers as a problem. It should. It is usually the single clearest signal that the group is not doing what it was funded to do, and it is almost never picked up by the metrics these programmes report.

This is the fourth-session cliff: the point at which a drop-in group stops being a group that is filling up and becomes a group that is failing, and nobody notices because the reporting was built to track headcount and satisfaction, not survival.

Why the usual metrics don’t catch it

Most social prescribing and community group evaluations lean on two numbers: attendance at each session, and a satisfaction or wellbeing score collected from whoever shows up. Both are easy to collect. Both are close to useless for spotting the cliff.

Attendance-per-session tells you how many people were in the room, not who they were. A group that loses seven of its original twelve members but replaces some of them with new drop-ins can report “steady attendance of eight” for months while actually retaining almost nobody. Satisfaction scores compound the problem: they are collected from the people still in the room, which means they systematically exclude the people the group failed to hold. A qualitative meta-synthesis of social prescribing found that participants describe benefit in terms of restored purpose and meaningful participation, not just contact — but that finding comes from people who stayed engaged long enough to be interviewed. The people who left after two sessions are not in that sample, and their absence is exactly what a satisfaction score cannot show.

The result is a report that reads well and a group that has hollowed out. Funders see stable numbers and high satisfaction; the programme team knows, informally, that it is really running for a small core of the same six people it always runs for. The evidence base largely reflects this same blind spot. A 2023 review of loneliness and isolation research found that inconsistent measurement across studies is one of the central barriers to comparing what works, and a 2025 systematic review protocol on social prescribing for older adults noted that only one peer-reviewed randomised controlled trial exists in the entire field. Programme-level reporting has the same weakness the research literature has: it counts what is easy to count, at the point in the programme when it is easiest to count it.

What to track instead

None of this requires more sophisticated instruments. It requires tracking the same people over time rather than the room.

1. Build a retention cohort from session one, not a running attendance count. Name the original attendees and track, session by session, what fraction of that specific group is still coming. This is the number that shows the cliff. A stable “eight attendees” is meaningless without knowing whether it is the same eight.
2. Set a checkpoint at session four, not session twelve. If retention of the original cohort has dropped below roughly half by then, treat it as a design problem to be diagnosed immediately, not a trend to monitor.
3. Ask leavers, not just stayers, why they left. A short call or text to anyone who attended once or twice and then stopped is more informative than another satisfaction survey of the core group. This is the population the evaluation is otherwise structurally blind to.
4. Separate the diversity of the group’s network from its size. A national survey of adults 45 and older found that the strongest predictors of loneliness were the size and diversity of a person’s social network, not attendance at any single activity. A group that retains six people who already knew each other is not building network diversity for anyone; it is a maintained friendship circle wearing a programme’s name.

What this means in practice: if your evaluation plan has one attendance line and one satisfaction score, add a retention-by-original-cohort figure before you add anything else. It is the cheapest number to collect and the one most likely to tell a funder something true.

Evidence status

Claim	Evidence status
Attendance counts alone obscure cohort turnover	Not directly studied, but a structural feature of how these numbers are calculated
Satisfaction surveys under-represent people who disengage early	Plausible and consistent with qualitative findings, not directly quantified
Structured, purposeful activity outperforms unstructured contact	Supported by qualitative meta-synthesis, though based on small, self-selected samples
Social prescribing reduces loneliness overall	Mixed — systematic reviews report positive impacts, but trial evidence is described as limited and heterogeneous
Volunteering-style engagement reduces loneliness in older adults	Supported by one randomised controlled trial, still a rarity in this literature

The last row matters for calibration. The Lancet Healthy Longevity’s dual randomised trial of volunteering among lonely older adults in Hong Kong is one of the few controlled tests of any loneliness intervention. Most of what practitioners rely on for programme design is uncontrolled evaluation, self-report from people who stayed engaged, and reviews that themselves flag heterogeneity as a limitation. That is not a reason to abandon drop-in groups. It is a reason to measure retention honestly rather than assume the model works because attendance held.

What this does not solve

Tracking retention will tell you when a group is failing. It will not tell you why people who never showed up in the first place stayed away, and it does nothing about reach. Drop-in groups recruit through GP referrals, link workers, posters, and word of mouth — channels that reach people already in some kind of contact with a service. People who are isolated precisely because they have no such contact remain outside the measurement entirely, cliff or no cliff. Fixing the fourth-session problem makes existing groups more honest about their own performance. It does not make them reach anyone new.