Evaluating Connection Training Without a Research Budget

You have run a training programme — for link workers, community health workers, employee resource group leads, whoever is supposed to help other people connect — and someone with the budget is now asking whether it worked. You do not have a control group. You do not have a grant for a trial. You have a spreadsheet, a training calendar, and maybe a follow-up survey template you found somewhere. This is the situation almost every training evaluation in this field actually starts from, and most of them respond by measuring the wrong thing at the wrong time.

The graduation-day survey

The default evaluation is a satisfaction survey handed out at the end of the last session. People rate the training highly because they are tired, relieved, and still feeling the social warmth of the room. This is the graduation-day survey, and it measures the training experience, not the training’s effect. It tells you nothing about whether the person’s practice changed once they were back at their desk facing a waiting list.

The systematic reviews of social prescribing — the closest adjacent field with a real evidence base — run into a version of this problem constantly. A qualitative meta-synthesis of how people experience social prescribing found that participants describe benefit extending beyond contact itself, toward restored participation and purpose, but that finding came from asking people well after the intervention, not on day one. The lesson transfers directly to training evaluation: what people say immediately after an intervention and what changes in their behaviour weeks later are different measurements, and only one of them is useful to a funder.

What you are actually trying to show

Before choosing an instrument, decide which of three different claims you are trying to support. They require different evidence and different timing.

1. Knowledge changed. Staff can now identify isolation risk factors, name the local referral pathways, or describe the distinction between loneliness and objective isolation — a distinction the literature treats as genuinely separate constructs with different correlates, not interchangeable terms.
2. Behaviour changed. Staff now ask about social connection routinely, refer at higher rates, or spend measurably different time on relational aspects of a case.
3. Downstream outcomes changed. The people staff work with report less loneliness, more social contact, or better wellbeing.

Most training evaluations quietly try to claim all three from a single post-training survey. That is the design error. Claim (1) is cheap to measure honestly. Claim (2) requires you to look at your own service data, not ask staff to self-report. Claim (3) requires something closer to the intervention research the field mostly does not have — the systematic review of social prescribing’s effect on loneliness found only nine eligible studies worldwide, and a 2025 review protocol on social prescribing for older adults notes that only one peer-reviewed randomised controlled trial exists in the entire area. If a controlled trial evaluating volunteering itself struggled to get built and run — the Hong Kong dual RCT on volunteering and loneliness among older adults is one of the only examples — a training programme evaluation with no dedicated budget has no business promising causal proof of downstream wellbeing change.

What this means in practice: size your evaluation claim to your evaluation budget. If you have no comparison group and no follow-up beyond eight weeks, you can credibly claim knowledge and behaviour change. You cannot credibly claim you reduced loneliness. Say so in the report rather than letting a funder infer more than the design supports.

A design you can run with what you have

Step 1: Pick a pre-existing measure over a bespoke one. Loneliness measurement is inconsistent across studies, which a 2023 review of the field flags as a barrier to comparing results at all. Where the outcome is client-facing, use the UCLA Loneliness Scale — the instrument the AARP Foundation’s national survey used, precisely because it is comparable to the wider literature rather than a one-off tool nobody else can validate against.

Step 2: Separate isolation from loneliness in your questions. The National Academies’ consensus report on older adults, and the clinical commentary that followed it, both push health systems toward routinely assessing isolation as distinct from loneliness. If your training is meant to change how staff assess clients, test whether staff can now correctly distinguish the two — a specific, checkable skill — rather than a vague “confidence in addressing loneliness” item.

Step 3: Measure behaviour from records, not recall. Pull referral counts, follow-up completion rates, or documented conversations about social connection from your existing case management system, for a period before training and a period after. This is the single highest-value thing a resource-constrained evaluation can do, because it does not depend on anyone accurately remembering or flattering their own practice.

Step 4: Time the follow-up at 8 to 12 weeks, not on the last day. This is long enough for the glow to fade and for staff to have hit real cases, short enough that you are not also capturing turnover and drift.

Step 5: Stagger the rollout if you train more than one cohort. You will rarely get a true control group, but if you train Team A in month one and Team B in month three, Team B’s pre-training data is a rough comparison for what “no training yet” behaviour looks like. It is not a randomised trial. It is still better than nothing.

Step 6: Report the limits explicitly. State what you measured, over what period, against what baseline, and what you did not measure. This is what separates a credible evaluation from a marketing document.

Evidence status of common training-evaluation claims

Claim	Evidence status
Staff can gain measurable knowledge of isolation vs. loneliness distinctions through training	Reasonable to claim; supported by consensus reporting that the two are separable constructs worth assessing differently
Training changes referral behaviour	Plausible and measurable from your own records; not established by any large body of independent evidence
Training reduces client loneliness	Not established; the underlying intervention evidence base for social prescribing itself remains thin, with review authors flagging a near-total absence of trials
Structured, purposeful activity works better than contact alone	Supported, from qualitative synthesis of client experience, though based on perception rather than controlled comparison
A single self-report survey at the end of training tells you the training worked	Not supported; measures experience of the training, not its effect

What this does not solve

This design will get you a defensible internal evaluation, not a publishable one. It cannot tell a funder your training caused a population-level drop in loneliness — nobody’s evaluation of a training programme alone can currently make that claim honestly, given how few controlled trials exist in the adjacent intervention literature. It also says nothing about staff who never volunteer for training in the first place, or organisations that lack a case management system to pull baseline behaviour data from. If your evaluation depends on records that do not exist yet, the first year of “evaluation” is building the record-keeping, not measuring the outcome.