How Social Prescribing Schemes Actually Evaluate What They Do

A commissioner asks a social prescribing service to show it reduced loneliness. The service has a part-time link worker, a spreadsheet, and eight months of referral data. It does not have a control group, a validated baseline for every client, or a research budget. This is the normal starting condition for most connection-focused programmes, and it is worth designing an evaluation around that condition rather than around the one a university would run.

Social prescribing — the practice of a health or care worker referring someone to non-clinical community activity, groups, or support rather than, or alongside, medical treatment — has been running long enough, and been reviewed often enough, that there is now a published pattern in how services measure what they do. It is not a good pattern in the sense of methodological rigor. It is a useful pattern in the sense that it shows what practitioners settled on when reality intervened.

What social prescribing services actually measure

A 2021 systematic review in the International Journal of Environmental Research and Public Health looked across the published social prescribing literature for evidence of individual and community well-being effects. The headline finding practitioners should notice is not the direction of the effect — it was generally positive — but what the effect was measured in. The review reports increases in self-esteem and self-confidence as key outcomes, not reductions on a loneliness scale. That is a real difference in what is being counted, and it did not happen by accident.

A separate systematic review, published the same year in Perspectives in Public Health, looked specifically at loneliness outcomes across nine studies. All nine reported positive individual impacts. Three went further and reported reductions in use of GP appointments, emergency services, social work contact, or inpatient care. Again, notice what got measured: service utilisation, which sits in administrative data services already hold, alongside self-report change, which required asking people something.

A 2022 qualitative meta-synthesis in BMC Health Services Research adds the reason services drift this direction. Participants in social prescribing programmes describe the benefit as extending beyond social contact itself, toward restored meaningful participation and purpose. Structured, purposeful group activity reads to participants as more valuable than contact alone. If that is what the people receiving the service say the service did for them, a loneliness-scale score that only asks about frequency of contact is measuring the wrong layer.

Put together, three things emerge as the de facto measurement toolkit of a service that cannot run a trial:

1. A short self-report measure, usually confidence, self-esteem, or a wellbeing scale, taken at referral and at a defined follow-up point.
2. Administrative service-use data the organisation already collects — GP contacts, A&E attendance, missed appointments — read before and after referral, where access permits.
3. Structured qualitative return — not a satisfaction survey, but something closer to what changed and for whom, gathered in a way that can be coded and counted, not just quoted.

None of this is a substitute for a controlled trial. It is what a service with a link worker and a spreadsheet can actually produce, repeatedly, without external funding for evaluation.

The named failure mode: the borrowed-scale problem

The most common mistake in this space is picking up a validated loneliness instrument — the UCLA Loneliness Scale is the one used in the national surveys that give practitioners their benchmark figures, including the AARP Foundation’s 2018 survey of adults 45 and older — and administering it to twenty-five service users at intake and exit, then reporting the change as evidence of impact.

Call this the borrowed-scale problem. The scale itself is not the issue; it is well validated, and using the same instrument as the national literature is exactly what lets you say your users started more isolated than the national average. The issue is what happens next. Twenty-five people is too small a sample for a scale designed to detect population-level differences to produce a stable, defensible result. There is no comparison group, so a change over three months cannot be separated from regression to the mean, seasonal effects, or simply people who were already getting better referring themselves into the service at a low point. And a single instrument administered twice tells you nothing about the self-esteem, purpose, and service-use dimensions that the published reviews suggest matter more to participants than raw social contact.

The result is a number that looks precise — “average loneliness score fell by 4.2 points” — and is not defensible under any scrutiny past the first question. Funders who have seen a few of these numbers stop believing them, which is worse for the next programme manager than reporting nothing.

An evaluation design that survives scrutiny at this scale

Component	What it answers	What it cannot answer
Validated scale (e.g. UCLA Loneliness Scale) at intake and exit, same clients	Did self-reported loneliness change for people who completed the programme?	Whether the programme caused it, or who dropped out before exit and why
Administrative service-use data, before and after referral	Did contact with GP, A&E, or crisis services change?	Whether reduced contact reflects less need or reduced access
Structured qualitative interviews or coded case notes at exit	What changed for participants, in their own account — purpose, confidence, contact, or something else	Generalisability; this is depth, not breadth
Referral and completion rates by source and demographic	Who is reaching the service and who is dropping out	Why people who never referred themselves are absent from all of the above

The design that survives scrutiny combines rows one and three, always, and adds row two wherever data-sharing agreements allow it. Row three is the one most services skip because it is slower to run, and it is the one the qualitative meta-synthesis suggests carries the outcome participants actually experienced as valuable. Skipping it to save time is a false economy: it is the only row that tells a funder why the scale moved, which is what turns a number into a case for renewed funding.

What this means in practice: Do not evaluate on a loneliness score alone. Pair it with one structured qualitative return per cohort — a short set of coded exit interviews, not a satisfaction form — and pull whatever service-use data your data-sharing agreement already permits. If you can only afford one addition to the scale, make it the qualitative return, not a bigger sample.

Where the evidence itself runs out

A 2023 review in BMC Public Health mapping the state of loneliness and social isolation research names inconsistent measurement as a structural barrier across the field, not a problem specific to under-resourced social prescribing services. Studies use different scales, different time windows, and different definitions of loneliness versus isolation versus lack of support, which means even well-funded academic evaluations struggle to compare results across programmes. This is worth saying to a funder directly: a service reporting an imperfect but consistent measure across cohorts is not falling short of an academic standard that the academic literature itself has met. It is falling short of a standard the field has not yet agreed on.

The National Academies’ 2020 consensus report on isolation and loneliness in older adults, and the clinician-facing commentary that followed it, both push toward routine assessment inside health and care settings rather than one-off programme evaluation. That is a different ambition — building loneliness screening into ordinary clinical contact — and it depends on infrastructure most social prescribing services do not control. The UK’s 2018 loneliness strategy took a similar structural approach, embedding measurement into national statistics rather than leaving it to individual providers. Both point toward where measurement should eventually sit. Neither solves the problem facing a service that needs a defensible answer for a funding panel next quarter.

What this does not solve

This approach produces a credible local evaluation, not proof of causal effect, and not a result that generalises past the population the service actually reaches. Every social prescribing evaluation reviewed here draws on people who were referred and who completed the programme — by definition, the people already inside a health or care system, already willing to take up a referral, already able to attend. The AARP survey’s finding that predictors of loneliness include the size and diversity of a person’s network and physical isolation itself suggests the people hardest to reach with a referral-based service are exactly the ones least likely to appear in its intake data at all. No measurement design fixes a reach problem. It can only be honest about it.