How to Choose Outcome Measures a Commissioner Will Actually Accept

The first draft of most programme evaluations picks the outcome measure last, after the logic model, the theory of change, and the staffing plan are already locked. That is backwards. A commissioner reviewing a renewal application is going to ask one question first: does this number mean what you say it means? If the answer is unclear, everything else in the report is decoration.

This is a procedure for choosing measures that survive that question, not just measures that are easy to collect at week one and week twelve.

Step 1: Decide whether you are measuring isolation or loneliness

These are not the same thing, and funders increasingly know it. Isolation is structural — the size and frequency of someone’s social contacts. Loneliness is subjective — the gap between the contact someone has and the contact they want. A person can be surrounded by people and lonely; a person living alone can be perfectly content. Research on older adults specifically warns against collapsing the two, because the risk factors and the appropriate response differ.

If your programme increases contact frequency — a walking group, a shared-meal scheme, a befriending rota — you are plausibly moving isolation. Whether you are moving loneliness is a separate empirical question, and you should not assume the first causes the second.

Decision: name which one you are targeting before you pick an instrument. If it’s both, measure both, separately, and report them separately.

Step 2: Use a validated instrument, not a bespoke one

The UCLA Loneliness Scale is the standard against which most of the epidemiological literature is calibrated. When AARP surveyed adults 45 and older, it used the full 20-item UCLA scale rather than a bespoke question set, which is precisely why that survey’s numbers are comparable to the rest of the field. A commissioner who has seen other applications will recognise it, know its properties, and trust a change on it more than a change on a locally invented five-point “how connected do you feel” question.

Bespoke instruments are not worthless — they can capture something the standard scale misses — but they cannot be your primary outcome if you want the result to travel outside your own organisation. Use a validated scale as the headline number and a bespoke measure, if you need one, as a secondary.

Step 3: Pick a comparator, and be honest that you probably don’t have one

This is where most social prescribing evaluations fall apart under scrutiny. Systematic reviews of social prescribing for loneliness report positive effects across nearly every included study — but reviewers are also explicit that trial evidence is thin and heterogeneous, and a 2025 protocol reviewing social prescribing for older adults’ isolation found only one peer-reviewed randomised controlled trial in the entire field. Before-and-after change in a single cohort, with no control group, is the default design in this space. It is not nothing, but it cannot rule out regression to the mean, seasonal effects, or the fact that people who join programmes were often already on an upward trajectory.

If you cannot run a control arm — and most local programmes cannot — say so in the report rather than letting the reader infer a causal claim you haven’t earned. A funder who has seen this failure mode before will trust you more for naming it than for hiding it.

Step 4: Set a realistic effect size expectation

Do not promise a transformation. The strongest evidence available on interventions that were actually tested against a control shows modest, specific effects. A randomised trial of befriending in residential aged care found a reduction of 2.39 points on the UCLA scale at eight weeks and 2.71 points at sixteen — real, but not dramatic. A larger trial testing telephone-delivered behavioural activation against a befriending control in older adults living in poverty and digitally excluded found the structured approach outperformed befriending at twelve months. Read together, these say two things: connection-focused interventions can move loneliness measurably, and befriending — the most commonly funded intervention — is not automatically the strongest one available.

Set your target reduction against this range. A commissioner who has seen these trials will be suspicious of a report claiming loneliness was “eliminated” or “dramatically reduced” for a light-touch programme running fewer weeks than the trials that produced modest results.

What this means in practice: state your target outcome as a specific, modest number on a named scale, against either a comparison group or an honestly labelled pre/post design. “Reduced UCLA Loneliness Scale scores by an average of X points over Y weeks, no control group” is a sentence a commissioner can trust. “Participants reported feeling much more connected” is not.

Step 5: Add one process measure the outcome measure can’t show

Qualitative work on social prescribing consistently finds that participants describe benefit beyond raw social contact — restored purpose, a reason to leave the house, a sense of being useful again — and that structured, purposeful activity seems to outperform unstructured contact. A loneliness scale won’t capture that. Attendance consistency, or a short question on perceived purpose or meaning in the activity, gives a commissioner a second signal when the headline number is ambiguous, which — given small sample sizes — it usually will be.

The evidence-status table

Claim	Evidence status
UCLA Loneliness Scale is the field standard for comparability	Well established
Isolation and loneliness are distinct and should be measured separately	Well established
Social prescribing reduces loneliness	Suggestive, but very few RCTs
Structured activity outperforms unstructured contact	Suggestive, qualitative evidence
Befriending is the most effective connection intervention	Contradicted — outperformed by structured approaches in head-to-head trials

What this does not solve

None of this fixes small sample sizes, short funding cycles, or the fact that the people who show up to be measured are already the ones who sought help. A rigorous outcome measure on a self-selected group of twenty people who opted into your programme tells you something real about them and very little about whether the same approach would work on the people who never walked through the door. Choosing the right instrument makes your evidence honest. It does not make it representative.