Evaluating a Borrowed Model: What Changes When You Move It to a New Population

You have a model that worked. A befriending scheme cut isolation among older adults; a volunteering programme moved the needle on loneliness in a trial population. Now someone wants to run the same thing for new parents, for remote employees, for university students. The instinct is to keep the manual and swap the participants. That instinct is the problem.

Interventions built for one population encode assumptions about that population’s causes of disconnection, its available time, its mobility, and what counts as a credible outcome. Change the population and those assumptions may no longer hold, even if the activity looks identical on paper.

Why age is the wrong thing to hold constant

Most of the loneliness intervention evidence base was built on adults 65 and older, because that is where the health system placed its attention first. The National Academies’ 2020 consensus report is explicit that roughly a quarter of older adults are socially isolated, and it frames the health care system’s role around identifying and referring this group. The one randomised trial in this literature with any real design rigor — the HEAL-HOA study, testing volunteering against a control among lonely older adults in Hong Kong — was built around older adults’ specific relationship to purpose and contribution, not a generic dose of “social contact.”

Move that model to college students and the mechanism may not transfer. A 2025 study of college students coming out of the pandemic found that what predicted wellbeing was not volunteering-as-purpose but the regularity and quality of specific connections. Move it to remote employees and you are dealing with a different construct altogether: research on healthcare workers found that workplace isolation and loneliness are separate constructs with different correlates, meaning a fix for one may do nothing for the other. Gallup’s 2024 workplace data shows fully remote employees reporting loneliness at 25% against 16% for on-site staff — a gap plausibly driven by structural absence of contact, not by the deficits in social skill or network size that older-adult programmes are often designed around.

The distinction between isolation (an objective lack of contact) and loneliness (a subjective feeling) matters more, not less, when you change populations. A 2024 study in Scientific Reports found the relationship between the two varies by age group. A programme built to fix isolation in an older population — get people out of the house, into a group — may land on a younger population that is not objectively isolated at all, just unsatisfied with the quality of contact it already has. Same activity, wrong target.

The adaptation checklist

Before running a borrowed model on a new group, work through these five questions. Skipping any one of them is how a programme ends up with attendance but no outcome movement.

1. What was the original population’s specific deficit — contact, quality, or purpose? The AARP Foundation’s 2018 survey of adults 45 and older found the strongest predictors of loneliness were network size, network diversity, and physical isolation — a contact-and-diversity problem. That is a different deficit from what a lonely new parent or an isolated remote worker typically reports.
2. Does the new population have the same practical constraints? Mobility, working hours, caregiving responsibilities, and access to transport all shape whether a weekly in-person group is even reachable. A model built for retirees with flexible daytime hours will not survive unmodified contact with shift workers.
3. Is the mechanism of change portable? Purpose and contribution (volunteering) worked in a trial of lonely older adults. Quality and regularity of existing ties mattered more for post-pandemic college students. These are not interchangeable levers.
4. What outcome measure actually applies? Isolation and loneliness need separate instruments and separate theories of change; conflating them in your evaluation plan will hide which one, if either, moved.
5. What does the evidence base actually cover? Be honest in your funder-facing documents about which claims are borrowed from a different population and which are supported for the group you are actually serving.

Evidence status across the adaptation

Claim	Evidence status for older adults	Evidence status when moved to a new population
Purpose-based activity (e.g., volunteering) reduces loneliness	Supported by one randomised trial (HEAL-HOA, older adults, Hong Kong)	Untested; mechanism may not generalise
Structured group activity outperforms unstructured contact	Reported qualitatively across social prescribing studies	Reported qualitatively; not population-specific evidence
Isolation and loneliness are distinct and need distinct measures	Established across the older-adult literature	Established in principle, direction of relationship varies by age group
Network size and diversity predict loneliness	Strong survey evidence, ages 45+	Plausible but not established for younger or working populations
Social prescribing improves self-esteem and confidence	Reported in a systematic review, mixed populations	Same review is not age-stratified; treat as general, not population-specific

The social prescribing literature itself is a useful caution here. A 2021 systematic review found consistent gains in self-esteem and confidence but noted limited trial evidence and heavy heterogeneity across programmes — meaning even the “established” outcomes are established loosely, across a grab-bag of different populations and delivery models. A 2022 qualitative synthesis found that what participants valued was restored meaningful participation, not contact for its own contact’s sake — which supports adapting the purpose mechanism across populations more than it supports adapting the format.

What this means in practice: Do not evaluate an adapted programme against the original’s published outcomes. Build a new baseline for the new population, on the deficit that population actually reports, and treat the original evidence as a hypothesis to test, not a result to expect.

The transplant failure mode

Call this the transplant failure mode: a programme runs faithfully to its original design, achieves normal attendance, and shows no measurable change in the outcome it was funded against — because the outcome it was built to move does not describe the new population’s problem. It gets read, wrongly, as an implementation failure. Staff get blamed for poor delivery of a model that was never suited to the target group in the first place. The fix is not better delivery. It is checking, before launch, whether the borrowed mechanism matches the borrowed population’s actual deficit.

What this does not solve

None of this tells you how to build a mechanism from scratch for a population with no evidence base at all — new parents, informal carers, gig workers — where you are extrapolating from adjacent literatures because nothing more specific exists. It also does not solve the reach problem common to every model in this space: whatever you adapt will still mostly land with people who opted in, and evaluation built on volunteers cannot tell you what would happen with people who never sign up.