Reach Is Not a Warm-Up Act for Outcomes

A funder asks for evidence that your walking group reduces loneliness. You have pre- and post- UCLA Loneliness Scale scores for the 34 people who completed both. The scores dropped nicely. You send the slide. Nobody asks the question that actually matters: out of how many people who were told about this group, or eligible for it, did 34 show up and stay long enough to finish a second questionnaire?

That missing denominator is not a technicality. It is the difference between “this intervention works” and “this intervention works for the kind of person who was already going to come back.” Those are different claims, and they support different decisions.

Two different questions, not one blended metric

Outcome measurement asks: for the people who took part, did anything change? Reach measurement asks: who took part, compared with who could have, and who dropped out along the way?

Programme managers routinely treat reach as throat-clearing before the “real” data — attendance figures on one slide, outcome scores on the next, with no attempt to connect them. That ordering is backwards. Reach determines what your outcome data is even a measurement of.

Consider two programmes with identical loneliness-score improvements. Programme A recruited from a GP referral list covering everyone flagged as isolated in a practice population, and 40% of those referred attended at least one session. Programme B recruited through a sign-up sheet at a community centre already known for its social calendar, and word of mouth pulled in whoever was already inclined to join things. The improvement scores might look the same. What they mean is not. Programme A’s number describes something close to the population you were asked to serve. Programme B’s number describes people who were, on some dimension, already primed to benefit — a group the systematic review of social prescribing and wellbeing outcomes flags as a persistent confound, given how heterogeneous these programmes are and how thin the trial evidence still is behind the individual accounts of benefit.

The named failure mode: the responder-only sample

Call it the responder-only sample. It happens when a programme reports outcomes only for people who completed both a baseline and a follow-up measure, with no accounting for who was invited, who never enrolled, and who enrolled and vanished. The responder-only sample is not fraud. It’s what happens by default when you don’t design for it, because non-responders don’t fill in forms and therefore don’t show up in your dataset at all.

The problem compounds with loneliness interventions specifically, because loneliness itself predicts attrition. People who are more isolated are less likely to have the stable routines, transport, or confidence to attend a second session, let alone complete a second questionnaire. A responder-only outcome score is therefore systematically biased toward the less-isolated end of your intended population — the opposite of who the programme was meant to reach.

This is not a hypothetical concern invented for this article. The protocol for a 2025 systematic review of social prescribing for older adults notes explicitly that despite widespread adoption, the effectiveness evidence remains unclear, and that only one peer-reviewed randomised controlled trial exists in the area at all. Most of what passes for evidence is uncontrolled, responder-only, and short.

What each measurement actually tells you

Question you’re trying to answer	Measure this	Not this
Is the programme reaching the people it was designed for?	Enrolment rate against the eligible or referred population, broken down by the subgroups you care about (age, isolation risk, deprivation, referral source)	Total headcount at sessions
Is the programme retaining people once they start?	Attrition by session number, not just start vs. end	A single completion percentage
Does participation change anything for the people who stay?	Validated pre/post measure (UCLA Loneliness Scale is the standard) on the full enrolled cohort, with non-completers coded as missing, not excluded	Scores from completers only
Would a stranger’s word-of-mouth referral produce the same result as a structured one?	Compare outcome by referral pathway	Assuming referral source doesn’t matter
Is the intervention itself doing the work, or would any structured activity have done as well?	A control or comparison arm — rare, but not impossible, as the HEAL-HOA trials show	A single-arm before/after design

Evidence status: how solid is each claim behind this

Claim	Evidence status
Loneliness and social isolation predict mortality and other health outcomes	Strong. Multiple large meta-analyses, including Holt-Lunstad’s 2015 review, converge on this.
Social prescribing produces qualitative benefits participants value	Reasonably solid, but from meta-synthesis of qualitative accounts rather than trials — participants describe restored purpose and participation, not just contact.
Social prescribing reduces loneliness scores at population scale	Weak. The 2025 review protocol found only one randomised trial in the older-adult literature; most evidence is uncontrolled.
Structured activity (volunteering, behavioural activation) outperforms informal contact for reducing loneliness	Moderate and improving. Randomised trials — the volunteering-focused HEAL-HOA trial and its behavioural-activation-and-mindfulness counterpart — both beat a passive or befriending comparison.
Befriending reduces loneliness	Moderate. A 2025 randomised trial in aged care found real reductions on the UCLA scale — but when tested head-to-head against structured psychological approaches, befriending was the arm that lost.
Programme reach data is being systematically tracked and reported in this field	Weak. It is the exception, not the norm, across the sources available on social prescribing evaluation.

What this means in practice: before you report a single outcome number, report the denominator it came from. If you cannot state how many people were eligible or referred, how many enrolled, and how many completed each stage, you do not yet have an outcome result — you have a description of the people who stuck around. Build the enrolment and attrition tracking into the programme from day one, not as an evaluation add-on at the end.

A minimum reach-tracking setup that doesn’t require a research team

You do not need a randomised trial to get an honest reach picture. You need four numbers, tracked from the start:

1. The eligible or referred population — everyone who was told about the programme or flagged as a candidate for it, whatever the referral mechanism.
2. Enrolment — how many of those actually signed up, broken down by whatever subgroup matters to your funder (age, isolation score at referral, neighbourhood).
3. Attendance by session, not just start and end, so you can see where people drop rather than just that they dropped.
4. Outcome data reported against the full enrolled group, with non-completion treated as a result, not an exclusion.

None of this requires special software. It requires deciding, before the programme starts, that these four numbers will be collected whether or not they flatter the results.

What this does not solve

Reach measured against a referral list still only tells you about people who entered the referral system at all. It says nothing about the isolated people a GP never flagged, the older adult who moved neighbourhoods after their nearest community centre closed — a pattern the 2025 research on disappearing third places found concentrated in exactly the census tracts with the highest social vulnerability — or the person too withdrawn to respond to any invitation, structured or otherwise. Improving your reach numbers inside a referral pathway is a real and useful thing to do. It is not the same as reaching people the pathway never touches, and no amount of better tracking inside a programme changes who gets offered a place in the first place.