Why Loneliness Predictors Should Decide Who You Recruit as a Volunteer

A cross-national study published in the European Journal of Epidemiology on 20 January mapped who is lonely, where, and why, across countries with very different social and economic conditions. The predictors it identifies — age, income, household structure, health status — are not new individually. What matters for programme design is that they cluster differently by country and by group, which means the person a scheme needs to reach in one place is often not the person who walks into the volunteer induction session.

That gap is the recruiting problem. Most befriending and social prescribing schemes recruit volunteers through the channels that produce reliable, socially confident, time-rich people: retirement associations, faith groups, existing volunteer databases, word of mouth among people already active in community life. That is a sound way to fill a rota. It is a poor way to match the profile of who is actually isolated, because the predictors the EJE study lays out point toward younger, lower-income, and more socially disconnected groups than the volunteer pool tends to contain.

The volunteer mirror

Call this the volunteer mirror: a scheme ends up staffed by people who resemble each other, and who resemble the volunteer coordinator, more than they resemble the people the scheme was built to reach. It is not a hiring failure. It is what happens by default when recruitment relies on networks, because networks recruit people like the people already in them. Robert Putnam’s account of declining associational life describes the same mechanism working in reverse: the people with the thickest existing networks are the ones with spare capacity to volunteer, and they are disproportionately not the isolated.

The consequence shows up downstream, not at recruitment. A volunteer matched with a young, low-income man with a thin social network — one of the predictor groups the EJE study and the AARP survey of adults 45 and older both flag as high-risk — needs a different kind of preparation than a volunteer matched with an isolated widow in her eighties. Training that assumes the second case as the default produces volunteers who are unprepared for the first, and who quit when the match doesn’t go the way induction implied it would.

What this means in practice: before recruiting a single volunteer, profile the loneliness predictors in your actual referral population — not the national averages. If your social prescribing pathway is pulling in more men under 40 than your volunteer pool can currently serve, recruit and train for that group specifically, rather than running one generic induction and hoping the match works itself out.

What the evidence supports about volunteer-delivered contact

The case for volunteer-delivered befriending is stronger than the case for social prescribing broadly, which is worth saying plainly because the surrounding literature is often vaguer than it should be.

Claim	Evidence status
Befriending reduces loneliness scores in older adults in aged care	Solid — a randomised controlled trial found a 2.39-point reduction on the UCLA Loneliness Scale at 8 weeks, 2.71 at 16
Structured volunteering reduces loneliness among lonely older adults generally	Solid — the HEAL-HOA dual RCT in Hong Kong tested prosocial engagement against a control arm, one of the few controlled loneliness trials that exists
Purposeful, structured activity works better than unstructured contact	Moderate — a qualitative meta-synthesis of participant accounts, not a trial
Social prescribing reduces GP or emergency service use	Weak-to-moderate — three of nine studies in one systematic review reported it; heterogeneous designs
Matching volunteers to the demographic predictors most at risk improves outcomes	Untested — a 2025 systematic review protocol notes only one peer-reviewed RCT exists for social prescribing with older adults at all

That last row is the honest summary of where this leaves you. Nobody has run the trial that tells you whether deliberately recruiting volunteers to mirror the predictor profile of your referral population outperforms recruiting them the usual way. The argument here is inference from what is known about who is lonely, not a tested intervention design.

Three things to change this quarter

1. Pull your referral data by age, sex, income band, and living arrangement, and compare it to your current volunteer roster on the same variables. If a mismatch exists — and for most schemes it will — that is the recruitment brief, not a training footnote.
2. Recruit outside the networks that already work for you. Employer partnerships, further education colleges, and sports clubs reach younger and lower-income volunteers that faith groups and retirement associations do not. This costs more coordinator time per volunteer recruited; budget for it rather than treating it as a stretch goal.
3. Train differently for differently-predicted groups. A volunteer being matched with a socially isolated younger man needs preparation on different failure points — withdrawal, no-shows, defensiveness about needing the service at all — than one being matched with an isolated older adult with declining mobility. One induction session covering both badly serves both.

What this does not solve

None of this addresses why the highest-risk groups the EJE study and comparable predictor research identify are the hardest to recruit as volunteers in the first place — they are also disproportionately the people with least spare time, least financial security, and least existing social capital to draw on. A recruitment strategy can close the demographic mismatch between volunteers and referrals to some degree. It cannot manufacture volunteer capacity in populations that are time-poor and isolated themselves, and it does nothing for the people who never reach a referral pathway at all. The Surgeon General’s 2023 advisory made the same point about the loneliness response generally: the interventions that exist mostly reach people who are already inside a system that can offer them.