What to Measure When Your Connection Programme Runs on Volunteers

Most evaluation plans for volunteer-based connection programmes — befriending schemes, peer visiting, phone buddy services — measure the wrong end of the pipeline. They track whether the person being visited feels less lonely at three months. They do not track whether the volunteer who was supposed to visit them is still doing it. When outcomes disappoint, the usual response is to question the theory of change. The more common cause is quieter: the volunteer left in week six, a replacement took eleven weeks to arrive, and the “intervention” the participant actually received was four visits and a long gap, not the twelve-week programme on the logic model.

If volunteers are the delivery mechanism, the volunteer pipeline is part of what you evaluate. Not as a staffing footnote. As a variable that determines dose, and dose is usually what determines effect.

The vanishing volunteer problem

Call it that, because it is the single most common way these programmes underperform on paper. A volunteer signs up, is trained, is matched — and leaves within the first few months, often before a second match is even established. The participant’s “intervention” becomes stop-start: an initial burst of contact, a gap while a new volunteer is recruited and trained, then a restart with a stranger. On an evaluation form, the participant still shows as “enrolled for six months.” In practice they received something closer to six weeks of actual contact.

This matters because the intervention evidence for social connection programmes is already thin. The American Heart Association’s 2022 scientific statement on social isolation named the absence of controlled intervention evidence as the central research gap in the field, not the strength of the observational link — that link, between weaker social ties and mortality, is well established. The Lancet Healthy Longevity trial of volunteering among lonely older adults in Hong Kong is notable mainly because it is one of the few randomised trials of a loneliness intervention at all, rather than an uncontrolled programme evaluation. Most of what practitioners have to work with is small, uncontrolled, and heterogeneous — a 2021 systematic review of social prescribing found all nine included studies reported positive individual impacts, but the review protocol itself, updated as recently as July 2025, still describes the effectiveness of social prescribing for older adults as unclear, with only one peer-reviewed randomised controlled trial in the area.

Given that thin base, a programme that lets dosage vary wildly by volunteer turnover and then reports outcomes as if dosage were constant is not just imprecise — it is actively contributing to the noise that makes this literature hard to build on.

Three things to measure that most programmes don’t

1. Match duration and continuity, not just match count. A programme reporting “140 matches made” tells a funder nothing about exposure. Report median match duration, the proportion of matches that survived to the programme’s intended end point, and the number of participants who experienced a volunteer changeover. Treat changeover as an adverse event to be minimised and disclosed, not a routine operational detail.

2. Time-to-replacement. When a volunteer leaves, how long before a new one is recruited, trained, and matched? This gap is invisible in most reporting and is often the actual explanation for weak effects. If replacement regularly takes two months, your twelve-month programme is delivering something closer to ten.

3. Volunteer attrition by stage, not just overall. Lump-sum turnover figures (“30% annual attrition”) hide where the leak is. Break attrition into: dropped out during training, dropped out in the first three months of matching, dropped out after six months. These have different causes and different fixes. Early training dropout usually points to a recruitment-fit problem — people signed up for the wrong reasons or didn’t understand the commitment. Early-match dropout usually points to a matching or support problem. Late attrition after six months is often just life circumstances, and harder to design against.

What this looks like against a funder’s usual questions

What this means in practice: if a funder asks “did loneliness scores improve,” answer that question, but attach a second table showing dose actually delivered per participant. A modest effect delivered at full dose is more useful evidence than a strong effect delivered at partial, uneven dose — the second number will not replicate.

Claim	Evidence status
Weaker social relationships predict higher mortality risk	Well established — Holt-Lunstad’s 2015 meta-analysis found odds ratios of 1.26–1.32 for loneliness, isolation, and living alone
Structured volunteering reduces loneliness in older adults	Emerging, trial-supported — the HEAL-HOA randomised trial found benefit from prosocial engagement and volunteering, one of few controlled tests in this space
Social prescribing schemes generally improve self-esteem and confidence	Reported consistently in qualitative reviews, but built on small, heterogeneous, mostly uncontrolled studies
Match continuity/dosage drives outcome strength	Not directly tested in the published literature at scale; inferred from general dose-response logic and consistent with why intervention effects are hard to detect
Volunteer-delivered programmes as effective as staff-delivered ones	Not established — the comparison is rarely made explicit in published evaluations

That last row deserves a plain statement: almost no published evaluation of a volunteer-delivered connection programme runs a clean comparison against a staff-delivered equivalent. Programmes choose volunteers for cost and reach reasons, not because the model has been shown equivalent. Say this to funders directly rather than letting them assume otherwise.

Building the volunteer-side data into your evaluation plan

1. At recruitment, record why each volunteer signed up, in their own categories (retirement, career-related, personal loss, general interest). This lets you later test whether motivation predicts retention — a question worth answering once, so future recruitment can screen for it.
2. At training, record completion and time-to-first-match. A long gap between finishing training and being matched is a known driver of early volunteer drop-off; measure it so you can shorten it.
3. During matching, log every contact, not just whether contact happened. Frequency and duration matter more than existence — a qualitative synthesis of social prescribing participants found the benefit they described went beyond mere contact, toward restored meaningful participation and purpose. Structured, purposeful contact appears to do more work than contact alone. If your data can only say “a visit occurred,” you cannot distinguish a five-minute doorstep exchange from an hour of conversation, and those are not the same intervention.
4. At exit — for both volunteers and participants — record the reason. Exit reasons are the cheapest diagnostic data a programme collects and the most often thrown away.
5. Report dose alongside outcome, every time. If your primary outcome measure is a loneliness scale (the UCLA scale is standard and lets you compare against national surveys such as AARP’s, which found one in three US adults 45 and older report being lonely), report it next to median contact hours delivered per participant, not just enrolment duration.

The National Academies’ 2020 consensus report on isolation and loneliness in older adults, and the clinical commentary that followed it, both push for routine assessment of isolation within health and care settings — the same discipline applies to the volunteer relationship that is supposed to address it. If you would not tolerate a clinical trial that failed to record whether patients took the medication, do not tolerate a connection programme that fails to record whether the volunteer showed up.

What this does not solve

None of this fixes the underlying evidence problem: the field still lacks large controlled trials of volunteer-delivered connection programmes, and a 2023 review of the state of loneliness and isolation research points to inconsistent measurement across studies as a structural barrier to comparing anything. Tracking dosage and attrition better will make your own evaluation more honest and more useful to your own decisions. It will not, on its own, produce the kind of evidence base the field is missing. And it does nothing for reach: better-measured volunteer programmes still tend to recruit volunteers and reach participants who are already somewhat connected — people with the time, health, and social confidence to volunteer, and people already inside a referral pathway that flagged them as isolated in the first place. The people most cut off from any institution that could refer them stay outside the measurement entirely, however carefully you track the rest.