Designing a Loneliness Evaluation You Can Actually Run

You have a twelve-month grant, a part-time coordinator, a weekly group of somewhere between eight and twenty people, and a reporting deadline. You are not going to run a randomised trial. The question is not how to approximate one. It is what set of numbers you can genuinely collect that will let you, and your funder, make a better decision next year than you made this year.

Most evaluation guidance for community programmes answers a different question — how to demonstrate impact — and it produces evaluations that look like research and function like marketing. This framework is built for the constraint you actually have.

Start from the decision, not the outcome

Before choosing an instrument, write down which decision the evaluation informs. There are usually only three:

1. Renewal. Does the funder continue at the same level? The evidence bar here is lower than people assume, and it is mostly about reach and retention.
2. Referral confidence. Will a GP practice or link worker keep sending people? They need to know who benefits and who does not.
3. Internal iteration. What do you change about the format, the timing, the venue, the group size?

These need different data. A renewal case needs a defensible participation funnel and one credible outcome measure. Iteration needs attendance patterns and dropout reasons, and almost no outcome data at all. Trying to serve all three with one questionnaire is how you end up with data too thin for any of them.

Step 1: Use an instrument someone else validated

The single most common self-inflicted wound is the bespoke questionnaire. A 2023 review in BMC Public Health identified inconsistent measurement as a central barrier to comparing findings across the loneliness literature — and if your instrument is homemade, you have no comparator at all, not even a bad one.

Use the UCLA Loneliness Scale, in the three-item or twenty-item form. The AARP Foundation’s 2018 survey of 3,020 adults aged 45 and older used the twenty-item version specifically so its results sat alongside the academic literature; that survey found one in three of those adults lonely, which gives you a population reference point. The UK’s 2018 loneliness strategy embedded a standard set of loneliness questions into Office for National Statistics collection, which is why UK programmes can benchmark. The CDC’s 2024 surveillance report gives US practitioners federal estimates for the same purpose.

Pick one. Do not modify the wording. Do not drop items to shorten it.

Step 2: Baseline at referral, not at first attendance

If you baseline at session one, you have already lost the people who were referred and never came — and those people are the most important group in your dataset. Your first questionnaire should be administered at the point of enrolment or referral, before anyone has sat in a room.

The last-week baseline

Here is the failure mode that shows up in a majority of closing reports: at week twelve, someone realises no baseline exists, so participants are asked to rate how lonely they were before the programme started. The retrospective answers are consistently and predictably worse than a real baseline, because people reconstruct the past in a way that makes sense of the present. The resulting improvement is large, clean, and worthless.

You cannot fix this at the end. It is a week-zero decision, and it costs about forty minutes of admin.

Step 3: Build the funnel

Attendance data is nearly free and, for most programmes, more decision-relevant than outcome data. Track five numbers per cohort: referrals received, people who attended once, people who attended a fourth time, median sessions attended, and people still attending at the end. Record why people stopped where you can.

This funnel tells you whether you have a recruitment problem or a retention problem — two failures with entirely different fixes, routinely conflated in reports that only present outcomes for people who completed.

Step 4: Find a comparison you can actually get

Pre-post scores on completers will overstate your effect. Everyone reading your report who knows the field will discount it. So build in the strongest comparison your operating reality permits, in this order of preference:

• A staggered start. If you have a waiting list, you already have a comparison group. Baseline everyone at referral; the second cohort’s pre-programme scores are your control period.
• Dose-response within the cohort. Do people who attended ten sessions differ from those who attended three? Confounded by motivation, but informative.
• A population benchmark. Compare your baseline distribution to the CDC or AARP figures to show who you are reaching.
• Nothing, stated plainly. If you have no comparison, say so in the report rather than letting the reader assume you had one.

What this means in practice: If your programme has a waiting list, you have already been handed a comparison group for free, and the only thing standing between you and usable data is collecting baseline questionnaires at referral instead of at session one. Do that and your evaluation becomes several orders of magnitude more credible at no additional cost.

Step 5: Measure participation, not just contact

A 2022 qualitative meta-synthesis in BMC Health Services Research found that participants in social prescribing described benefit extending well beyond social contact — restored meaningful participation and a sense of purpose — and that structured, purposeful group activity appeared more effective than contact alone. If loneliness scores are your only outcome, you will miss the thing your participants say matters.

Add two or three items on purposeful activity and confidence. The 2021 systematic review in IJERPH identified self-esteem and self-confidence as key reported outcomes of social prescribing, so these are not fringe measures.

What the evidence will and will not carry

Claim	Evidence status
Social disconnection predicts early mortality	Strong. Holt-Lunstad’s 2015 meta-analysis: isolation OR 1.29, loneliness OR 1.26, living alone OR 1.32
Social prescribing improves self-reported loneliness	Suggestive. All nine studies in the 2021 Perspectives in Public Health review were positive, but few trials and heavy heterogeneity
Programmes raise confidence and self-esteem	Moderate, from a systematic review that itself flags limited trial evidence
Programmes reduce GP, emergency or inpatient use	Weak. Three of nine studies in that review reported reductions
Structured activity beats contact alone	Qualitative only
Intervention effects are established generally	No. The American Heart Association’s 2022 scientific statement named the absence of intervention evidence as the central research gap

Do not claim service-use reduction unless you have the data yourself. It is the claim funders most want and the one the literature least supports.

What this does not solve

None of this addresses reach. Your funnel will describe the people who arrived, and almost everyone who arrives at a connection programme either sought it out or was referred by someone already in contact with them. KFF’s 2024 survey linked experiences of discrimination to smaller social support networks — and those are disproportionately people your referral routes do not touch. An evaluation designed around enrolment cannot see them, and it will quietly report success while missing the population with the greatest need. Measuring who is absent requires a different exercise entirely, usually starting with the referral source rather than the programme.