Nine Studies, Nine Positive Results

The 2021 systematic review of social prescribing and loneliness published in Perspectives in Public Health identified nine eligible studies. All nine reported positive individual impacts. Three also reported reductions in GP, emergency, social worker, or inpatient service use.

Nine out of nine is the finding worth sitting with. Not because the studies are wrong, but because a literature in which every study works is a literature that has not yet been built in a way that would allow a study to fail. The companion review published a month earlier in the International Journal of Environmental Research and Public Health reported gains in self-esteem and self-confidence while stating plainly that trial evidence was limited and programmes were heterogeneous. Those two sentences describe the same problem from different ends.

This matters most for the newest part of the field: tools and platforms that organise in-person meeting. That category has produced almost no evidence at all, and it is worth being precise about why, and about what a negative finding would even look like if one arrived.

Three things a null result requires

A null result is not the absence of a positive one. It is a specific evidentiary object, and it needs three components.

First, a named instrument administered identically at both ends of the observation window — the 20-item UCLA Loneliness Scale, a three-item short form, an ONS-style direct question, something with a documented distribution. Second, a comparison condition: a waitlist, an active control, a matched non-joining group. Third, a threshold declared before data collection, so that a two-point movement can be called flat rather than argued over afterwards.

Strip any one of those out and “no effect” stops being a sayable result. Uncontrolled pre-post designs with self-selected participants and outcome measures chosen after the fact do not produce nulls, because almost nothing in them can come out flat. This is not a criticism of the nine studies. It is a description of what they were built to do.

The platform category cannot report a failure yet

Meetup has been running recurring in-person interest groups since 2002 and reports roughly 60 million members. That is two decades of the largest mainstream infrastructure for organising face-to-face gatherings, and the publicly reported metrics are members, groups, and events.

There is no published series of loneliness scores for people who joined a group and people who considered joining and did not. No before-and-after UCLA readings, no network-size measure, no comparison arm. Whatever the true effect of joining a recurring in-person group is — large, small, zero, or negative for some subgroups — the measurement that could reveal it is not being taken.

The consequence is that membership growth carries no information about loneliness at all. Cigna’s 2020 report, describing fielding conducted before the pandemic, put the share of U.S. adults reporting they sometimes or always feel lonely at 61%, up seven points year over year. Platform membership grew across the same period. Those two facts are mutually compatible in every direction, and neither constrains the other.

What a real negative finding would look like

The most likely shape of a genuine null in this area follows directly from the isolation–loneliness distinction. Holt-Lunstad’s 2015 meta-analysis reported an odds ratio of 1.29 for social isolation, 1.26 for loneliness, and 1.32 for living alone, with effects persisting after adjustment for health status. Those are three separate coefficients because they are three separate constructs. Isolation is a property of a network. Loneliness is a subjective state.

So the interesting negative result is not “nothing happened”. It is: a tool measurably increases contact frequency and network size, and loneliness scores do not move. That would be a structural success and a subjective failure, and it would be genuinely informative — it would tell programme designers that contact volume is the wrong target.

The qualitative meta-synthesis published in BMC Health Services Research in October 2022 points at roughly this boundary from the other side. Participants described benefit that extended beyond social contact into restored meaningful participation and purpose, and structured, purposeful group activity appeared more effective than contact alone. Read as a hypothesis, that implies contact-only interventions are the plausible null case. But it is derived from participant accounts, not from a randomised contrast between structured activity and unstructured contact. Nobody has run that comparison.

An empty cell is not a null result

The observational mortality literature is in a different position, and the difference is instructive. Holt-Lunstad’s 2010 meta-analysis pooled 148 studies and 308,849 participants and found a 50% increased likelihood of survival for those with stronger social relationships. The American Heart Association’s scientific statement of August 2022 put social isolation and loneliness at roughly 30% increased risk of heart attack, stroke, or death from either — 29% for heart attack or death from heart disease, 32% for stroke.

That statement also did something the intervention literature has not managed: it named the absence of intervention evidence as the central research gap in the field. An identified gap is an empty cell, not a null. The distinction is easy to blur in public discussion, where “no evidence that it works” and “evidence that it does not work” get used interchangeably. On connection interventions delivered or organised through digital tools, the field is firmly in the first state.

Prevalence surveys report levels, and levels cannot be null

AARP’s 2018 survey of 3,020 adults aged 45 and older used the 20-item UCLA scale and found one in three lonely. Within that sample, 33% of respondents who had spoken with their neighbours reported loneliness against 61% of those who never had. Harvard’s Making Caring Common report, published in February 2021, put serious loneliness at 36% of Americans and 61% of young adults aged 18 to 25.

These are cross-sectional levels. The AARP neighbour contrast is a 28-point difference with no established direction — lonelier people may talk to neighbours less, or vice versa, and a single wave cannot separate them. There is no null available here because there is no effect being estimated.

Repeated administration of the same instrument is the one route to a flat finding in survey work, which is what makes Cigna’s seven-point year-over-year change unusual. It is also worth asking whether a year of no movement would have been written up and released at all.

The one cleanly falsifiable claim in the area

Dunbar’s number is the exception. The figure of about 150 was derived by fitting a relationship between neocortex ratio and group size across primates and inserting the human value, which produced 147.8, and the account Dunbar published in 2021 emphasises the layered structure — roughly 5 intimates, 15 close friends, 50 friends, 150 meaningful contacts — as much as the headline number.

That is a quantitative prediction about ceilings, and platform-mediated networks are a natural test of it. A tool that durably doubled a user’s 15-person layer over several years would be a disconfirmation. So would evidence that the layers scale with platform affordances rather than holding constant across them. The prediction is precise enough to lose, which is more than can be said for most claims in this literature.

The third-place and social-infrastructure arguments — Oldenburg’s in 1989, Putnam’s account of declining civic participation and informal socialising in 2000, Klinenberg’s case in 2018 that libraries, parks, and shared physical space measurably shape contact rates — are harder to falsify, because exposure to social infrastructure is never assigned. A negative finding would have to look like a jurisdiction that closed its shared spaces and showed no subsequent change in contact rates, and nobody sets that up deliberately.

What would move this forward

A registered trial of a single meeting tool, with a waitlist control, a pre-specified loneliness instrument, a separate network-size measure, a minimal important difference declared in advance, and a stated commitment to publish the result whichever way it falls.

The strongest claim in the field is Holt-Lunstad’s 2021 argument that social connection is a modifiable protective factor belonging alongside diet, exercise, and smoking in preventive frameworks. Modifiable is the operative word, and it is an intervention claim. It currently rests on a body of studies that has not once reported a failure — which, at nine out of nine, is closer to a measurement artefact than a track record.