What Prevalence Numbers Cannot Tell Policymakers

The U.S. Surgeon General’s 2023 advisory states that approximately half of American adults experience loneliness. Gallup’s 2023 survey of 142 countries puts the global figure at 24% feeling very or fairly lonely. The Harvard Making Caring Common survey found 36% of Americans reporting serious loneliness, rising to 61% among adults aged 18 to 25. These numbers get cited interchangeably in policy documents and news coverage, as though they are three measurements of the same underlying quantity that happen to disagree. They are not. They come from different instruments, asked of different populations, at different points during and after the pandemic. The gap between 24% and 61% is not noise to be averaged away. It is information about what each survey actually asked.

That distinction matters more than it sounds like it should, because a great deal of policy now rests on the premise that loneliness is measurable with enough precision to track, target, and evaluate an intervention against. The evidence supports the first two. It does not yet support the third.

Prevalence is not one number

The AARP Foundation’s 2018 survey of adults 45 and older used the 20-item UCLA Loneliness Scale, a validated academic instrument, and found one in three reporting loneliness. That figure is directly comparable to decades of clinical research because it uses the same tool. The Gallup-Meta State of Social Connections survey, fielded across 142 countries from June 2022 to February 2023, asked a single-item question about feeling socially connected, and found 72% answering yes. Gallup’s companion analysis found connectedness and loneliness are not simple inverses of each other; a person can report both. Harvard’s Making Caring Common survey, fielded near the start of the pandemic, asked about “serious loneliness” without the UCLA instrument’s multi-item structure.

None of these studies is wrong. They are answering different questions with different tools, and the resulting numbers are not the same currency. A national strategy that cites “half of Americans are lonely” in one paragraph and “a third of older adults are lonely” in the next, without noting the instruments differ, is not describing a trend. It is describing a measurement artifact dressed as a trend.

The BMC Public Health review of the state of loneliness research, published in mid-2023, names this directly: inconsistent measurement across the field is a structural barrier to comparing findings, not an incidental inconvenience. A field where the prevalence estimate swings by a factor of two depending on the instrument used is a field where “loneliness went up” or “loneliness went down” claims require the underlying question text before they mean anything.

What the mortality data actually licenses

The strongest empirical result in this literature is not a prevalence figure at all. It is Julianne Holt-Lunstad’s 2015 meta-analysis, which found social isolation carries an odds ratio of 1.29 for early mortality, loneliness 1.26, and living alone 1.32, holding after adjustment for baseline health. The American Heart Association’s 2022 scientific statement, led by Crystal Cene, converged on similar magnitudes for cardiovascular outcomes specifically: roughly 29% increased risk of heart attack or death from heart disease, 32% increased risk of stroke.

These are association estimates from observational cohorts. They establish that social disconnection is a real, replicated, dose-relevant risk factor, on the order of other well-known ones. What they do not establish is that intervening on connection changes the outcome. The AHA statement says this plainly: it identifies the absence of intervention evidence as the central research gap in the entire field. A risk factor identified through observational epidemiology is a legitimate basis for concern. It is not, by itself, a basis for predicting that a specific program will reduce mortality, because the studies were not designed to test that.

The social prescribing gap

This is where the evidence base thins fastest, and where policy has moved furthest ahead of it. Social prescribing, the practice of a clinician referring a patient to a community activity or group rather than a treatment, has become a preferred policy response in the UK and elsewhere. Two systematic reviews from 2021 looked at what evidence exists. One found all nine included studies reported positive individual-level effects on loneliness. The other, examining wellbeing outcomes more broadly, found gains in self-esteem and self-confidence but noted limited trial evidence and substantial heterogeneity across programs.

Reported positive effects across small, heterogeneous, largely uncontrolled studies is a genuinely different claim from evidence that social prescribing reduces loneliness at population scale, and the two get treated as equivalent in a lot of secondary reporting. Neither review found a randomized controlled trial with a comparison group large enough to rule out the possibility that people who accept a referral and attend are already the people most likely to improve regardless of the referral.

What better evidence would require

Three separate gaps, all obscured when prevalence figures are the headline. First, comparability: an instrument used consistently across time and geography, the way the UCLA scale allows within academic research but government surveys largely do not adopt. Second, a causal design: a trial that randomizes people to an intervention and a genuine control, tracks the same validated loneliness measure at multiple points, and follows for long enough to observe whether an effect persists once novelty fades. Third, an accounting of who does not show up. Social prescribing evaluations study people who accepted a referral and attended sessions; they say very little about the substantial number who were referred and did not go, who may be the group at greatest actual risk.

None of this argues against acting on the observational mortality evidence, which is robust enough on its own terms to justify treating disconnection as a public health concern. It argues against letting a prevalence statistic and an intervention’s plausibility substitute for evidence that the intervention works. The Surgeon General’s advisory frames loneliness as comparable in mortality risk to smoking up to 15 cigarettes daily. Tobacco control had decades of trial evidence on cessation methods before it had a national strategy built on that comparison. Social connection policy currently has the risk-factor evidence and not yet the intervention evidence, and the two should not be described as though they arrived together.