When the Ask Outruns the Evidence

A commissioner asks for a business case showing that a new befriending scheme will reduce GP attendance and cut mortality risk among isolated older residents. You know the loneliness literature. You know Julianne Holt-Lunstad’s meta-analyses put weak social relationships on a par with smoking and obesity as a mortality risk factor. So you write the case, cite the 2010 and 2015 papers, and promise the commissioner a reduction in death.

This is the single most common evidence error in social connection programme design, and it happens because the underlying science is genuinely strong. The mortality findings are not the weak link. The weak link is the leap from “loneliness predicts mortality across a population of 300,000 people” to “this 40-person befriending pilot will move that number.” Nobody has run the trial that would justify that sentence, and nobody is going to, because a mortality-powered RCT of a community programme would need years and a sample size no local pilot has.

The mortality slide

Call this the mortality slide: taking a population-level association and re-badging it as the expected effect of a specific intervention. It happens in almost every funding bid in this sector because the association evidence is the most quotable thing available and program budgets need a hook. Holt-Lunstad’s 2010 review, covering 148 studies and 308,849 participants, found that stronger social relationships were associated with a 50% increased likelihood of survival. That is a real, well-replicated finding about the value of social connection as a health factor. It is not a finding about what any particular programme does to mortality, and treating it as one sets a promise no delivery team can keep and no evaluator can verify.

The tell that you’re mid-slide: the citation in your bid is a meta-analysis of observational cohort studies, and the claim in your bid is about an intervention’s causal effect. Those are different categories of evidence, and funders rarely ask you to distinguish them, which is exactly why practitioners get away with blurring them until an evaluation comes due and nothing measurable has moved.

Three tiers, not one pile of evidence

Treat every claim in a bid or logic model as belonging to one of three tiers, and say which tier out loud.

Tier 1 — mechanism. Why connection matters at all. This is where Cacioppo’s framing of loneliness as an aversive physiological signal sits, alongside Oldenburg’s account of third places as the informal infrastructure of social contact and Klinenberg’s argument that shared physical space measurably shapes rates of contact. This tier explains why you’d bother designing a programme in the first place. It does not tell you your programme will work.

Tier 2 — association. What is correlated with what, at population scale. Holt-Lunstad’s mortality figures live here, as does the National Academies’ 2020 finding that roughly one quarter of adults 65 and older are socially isolated. Strong, well-replicated, and directly relevant to why the problem is worth funding. Still not evidence that your specific delivery model changes the outcome.

Tier 3 — intervention. What happens when you actually run something. This is thin, and where it exists it is mixed. A 2021 systematic review of social prescribing found reported increases in self-esteem and confidence, but flagged limited trial evidence and heterogeneity across programmes. A separate systematic review of social prescribing and loneliness found all nine included studies reported positive individual impact, but only three reported any reduction in service use — GP contact, emergency attendance, social worker or inpatient use. That is the actual, current evidence base for the claim commissioners most want: that connection programmes save the health system money. It is thinner than most bids imply, and it does not support a mortality claim at all.

The fix is not to stop citing Tier 1 and Tier 2 evidence — it is legitimate and it is why the sector exists. The fix is to never let it stand in for Tier 3 evidence in the sentence that promises an outcome.

What to do when the ask sits above the evidence

1. Name the tier the ask is really asking for. If a funder wants “reduced hospital admissions,” that is a Tier 3, intervention-specific claim, and you need to say plainly that the sector-wide trial evidence for it is thin and mixed rather than borrow a Tier 2 statistic to cover the gap.

2. Offer the claim the evidence will actually support. Social prescribing reviews consistently support claims about self-esteem, confidence, and self-reported loneliness reduction. Those are legitimate, evidenced, and measurable with instruments already validated in the literature, including the UCLA Loneliness Scale used in the AARP national survey. Pitch those, not the downstream health-economics claim nobody has demonstrated at programme scale.

3. Separate the rationale slide from the outcomes slide. Put the mortality and isolation-prevalence statistics in the section that explains why the problem matters. Put only intervention-level, programme-specific claims in the section that says what your funding will produce. A funder who sees both in the same slide will conflate them even if you don’t intend them to.

4. If a funder insists on a Tier 3 claim you can’t evidence, put a number and a caveat on it, not a promise. “Comparable social prescribing programmes have reported reduced self-rated loneliness; reductions in service use were reported in a minority of studies reviewed” is defensible. “This programme will reduce emergency admissions” is not, and it will be the line the next evaluation is measured against.

5. Build the evaluation around what Tier 3 evidence already exists for, not around what the rationale slide claimed. If you promised mortality reduction, you have committed to an evaluation you cannot deliver on any realistic timeline or budget. If you promised measured change in loneliness scores and self-confidence, you have committed to something achievable within a single funding cycle.

What this means in practice: Draw a hard line between the evidence that justifies funding a category of programme and the evidence that predicts what your specific programme will do. Cite the mortality and isolation statistics to make the case for why the problem is worth money. Cite only the intervention-level reviews — and their caveats — when promising an outcome. If a funder asks you to promise something in the second category that only the first category supports, say so in the bid, in writing, before the money moves.

Evidence status by claim

Claim	Evidence status
Weak social ties predict higher mortality risk	Strong — large meta-analytic base (Holt-Lunstad, 2010 and 2015)
Isolation and loneliness are common among older adults	Strong — national-scale consensus finding (National Academies, 2020)
Third places and shared physical infrastructure shape rates of informal contact	Theoretical and observational, not causally tested at intervention scale
Social prescribing improves self-esteem and confidence	Moderate — consistent finding across a systematic review, though trial quality is limited
Social prescribing reduces loneliness	Moderate — positive in all nine studies in a systematic review, but the review notes heterogeneity
Social prescribing reduces GP, emergency, or inpatient service use	Weak — reported in only three of nine studies reviewed
A specific local connection programme reduces mortality	No direct evidence exists at this scale; do not claim it

What this does not solve

This framework tells you how to avoid promising more than the literature supports. It does not fix the underlying problem, which is that funders often want the Tier 3 claim precisely because it is the only one that justifies the spend to their own board. Refusing to make the mortality claim may cost you the bid. That is a genuine trade-off, not a technicality, and no amount of careful evidence-tiering makes it disappear. The most honest version of this guidance is that some funding will go to whoever overstates the evidence with more confidence than you did.