Choosing Outcome Measures a Commissioner Will Accept

A commissioner reading your funding application will ask one question before any other: how will you know if this worked? Get the answer wrong and it doesn’t matter how good the programme is — the money goes somewhere with a cleaner evaluation plan. Get it right and you have bought yourself two or three years of runway. This is a guide to getting it right, and it starts from a fact most programme design documents skip past: the loneliness and isolation literature does not agree on how to measure its own subject.

A 2023 review of the state of loneliness and social isolation research found inconsistent measurement to be a structural barrier to comparing findings across studies, not a minor footnote. If the researchers can’t agree, you cannot assume your commissioner’s preferred metric is the obviously correct one. You have to choose, and you have to be able to defend the choice.

Isolation and loneliness are not the same thing

This distinction does the most work of anything in this article, and it is the one most programme documents blur. Social isolation is an objective, structural condition — how many people you see, how often, how varied your network is. Loneliness is the subjective distress of feeling short of connection, and a person can be isolated without feeling lonely, or surrounded by people and feel lonely anyway. The National Academies’ 2020 consensus report on older adults treats these as distinct constructs requiring different measurement approaches, and its clinical commentary pushes the same point further: health systems need routine assessment tools that don’t conflate the two.

This matters for outcome selection because your intervention probably changes one more than the other. A befriending scheme that pairs someone with a weekly visitor changes structural contact — that’s isolation. A group programme built around shared purpose and identity is aiming at the subjective sense of belonging — that’s loneliness. If you pick a loneliness scale to evaluate a pure contact-frequency intervention, you may show no effect and conclude the programme failed, when in fact you measured the wrong thing.

What the evidence actually supports

Claim or measure	Evidence status
UCLA Loneliness Scale (or short forms) as the primary outcome	Strong — the standard instrument in the academic literature; AARP’s 2018 and 2025 surveys used it specifically so results would be comparable to research literature, not just to each other
Reduced GP, A&E, or social care contact as a programme outcome	Moderate — three of nine studies in a 2021 systematic review of social prescribing found this, but the review notes small samples and inconsistent follow-up
Self-esteem and confidence gains from social prescribing	Moderate — reported consistently in a 2021 systematic review, but self-report, uncontrolled, and short follow-up in most included studies
Mortality or cardiovascular risk reduction as a stated outcome for a single programme	Weak to absent — Holt-Lunstad’s 2015 meta-analysis establishes the population-level association between isolation and early mortality, but no local programme evaluation can credibly claim to move this number; use it in your rationale, never in your outcomes
“Reduced loneliness” measured only pre/post with no comparison group	Weak — the effect could be regression to the mean, seasonal, or attributable to anything else happening in a person’s life that quarter
Structured, purposeful group activity outperforming unstructured social contact	Moderate-to-strong — a 2022 qualitative meta-synthesis found participants describe benefit tied to restored purpose and meaningful participation, not contact alone; a 2024 RCT of volunteering against a control found similar
Befriending as an active-control comparator that a purpose-built intervention needs to beat	Increasingly strong — a 2025 RCT found befriending reduced UCLA scores by real, measurable amounts, and a 2026 trial found a structured behavioural-activation intervention beat a befriending control at 12 months. Befriending works. It is simply not always the best available option, and commissioners are starting to know that.

The pattern across this table is consistent: self-report perception measures are on firmer ground than service-use proxies, and comparison to an active control is far more persuasive than a before-and-after number with nothing to compare it to. A 2025 systematic review protocol on social prescribing for older adults notes that only one peer-reviewed randomised controlled trial exists in the entire field despite the growing volume of programme spend. That gap is precisely why a commissioner who has seen several proposals will treat an uncontrolled pre/post loneliness score with polite suspicion.

The named failure mode: the good-news-only baseline

This is what happens when a programme measures loneliness only among people who complete it. Attrition in connection programmes is not random — people who feel worse, or whose lives got harder, tend to drop out before the endline survey. What’s left is a completer sample that looks like it improved, when the honest picture is that the programme retained the people it was already working for and lost the people it wasn’t. A commissioner who has read one evaluation report before yours will ask about your dropout rate and whether you measured anyone at baseline who didn’t finish. If you don’t have an answer, the good-news-only baseline is what they will assume happened.

The fix is not complicated, just unglamorous: report attrition, report baseline loneliness scores for dropouts versus completers where you can, and don’t claim an effect size derived only from finishers.

A sequence for choosing measures

1. Name the mechanism before the metric. Is the intervention increasing contact frequency, network diversity, subjective belonging, or access to a specific relationship type (a befriender, a peer, a mentor)? Choose the outcome that maps onto that mechanism, not the one that sounds most impressive.
2. Use a validated scale, not a bespoke one. The UCLA Loneliness Scale or its short forms let a commissioner compare your result to a body of existing evidence. A programme-specific five-question survey designed in-house cannot be benchmarked against anything, however well-intentioned.
3. Build in a comparator, even an imperfect one. A waitlist group, a lower-intensity version of the same programme, or simply comparing completers against a matched group of non-participants is more credible than nothing. The 2026 behavioural-activation trial is instructive precisely because it compared against befriending rather than against no intervention at all — that’s the standard the field is moving toward.
4. Decide your timepoint based on what the mechanism predicts, not administrative convenience. The 2025 befriending trial in aged care found effects at 8 and 16 weeks; the 2026 behavioural-activation trial measured at 12 months. If your programme’s theory of change says effects should build slowly, don’t measure at week six and report null results.
5. Keep service-use and cost outcomes secondary, not primary. They are what funders often want most, but the evidence connecting a social connection programme to a specific reduction in GP contacts is thinner than the evidence for self-reported loneliness change. Offer it as a secondary, exploratory outcome, and say so.

What this means in practice: lead your evaluation plan with a validated self-report loneliness or isolation measure matched to your programme’s actual mechanism, report it against a comparator group, disclose attrition honestly, and treat service-use reductions and cost savings as secondary claims you are testing, not promises you are making.

The UK’s 2018 loneliness strategy and the U.S. Surgeon General’s 2023 advisory both pushed loneliness measurement into routine data collection at a national level. That has raised the bar for local evaluation, not lowered it — commissioners increasingly have a national baseline to compare your numbers against, and a submission that ignores that baseline looks parochial by comparison.

What this does not solve

Choosing the right outcome measure does not fix a weak intervention, and it does not fix the reach problem that sits underneath almost all connection programming: the people you can measure are the people who showed up, completed intake, and stayed engaged long enough to be surveyed at the endpoint. Everyone the programme never reached — the isolated person who never heard about it, the lonely person too depleted to attend an information session — is invisible in every metric on this list, however well you choose it. A good outcome framework tells a commissioner whether the programme worked for the people it served. It says nothing about the much larger group it never touched.