The 50 Percent Figure: How One Meta-Analysis Became a Policy Talking Point

One number turns up in nearly every policy document on loneliness published in the last decade: people with stronger social relationships have a 50% greater likelihood of survival than those without. It appears in the UK government’s 2018 loneliness strategy, in the National Academies’ 2020 consensus report on older adults, in Vivek Murthy’s 2020 book, and in dozens of press releases that cite none of the above but repeat the figure anyway. The number comes from a single 2010 meta-analysis by Julianne Holt-Lunstad and colleagues, published in PLoS Medicine. It is worth tracing what that number actually measured, because the version now circulating in policy prose has drifted some distance from it.

What the 2010 paper measured

Holt-Lunstad’s meta-analysis pooled 148 studies covering 308,849 participants and asked a specific question: across studies that tracked social relationships and later mortality, what is the average increase in the odds of surviving over the follow-up period for people with stronger versus weaker social relationships? The pooled answer was an odds ratio corresponding to a 50% increased likelihood of survival. That is a summary statistic across a heterogeneous set of studies, using varied measures of social relationship — some looked at network size, some at marital status, some at participation in social activities — and varied follow-up periods, often years.

Two things about that structure matter for what happened next. First, it is a pooled effect across observational studies, nearly all longitudinal but not experimental. The paper can speak to association over time, not to what happens if a particular intervention increases someone’s social contact. Second, “50% greater likelihood of survival” is a population-level summary of relative odds, not a probability that applies uniformly to a given person. It says something true and useful about study populations. It is a poor sentence to hand to an individual reader trying to interpret their own risk.

The comparison that made it travel

What made the figure spread as fast as it did was a secondary claim also drawn from Holt-Lunstad’s work: that the mortality risk associated with weak social relationships is comparable in magnitude to other well-established risk factors — the kind of comparison that made “loneliness is as bad as smoking” a stock phrase in press coverage of the loneliness field, even though no single supplied source here states that exact smoking comparison. What the 2010 paper does establish, and what is defensible to say, is that the survival effect size sits in the same range as other recognized mortality risk factors. That is a comparison of pooled effect sizes across different literatures, not a claim that loneliness and smoking operate through the same mechanism or that reducing loneliness would reduce mortality by a proportional amount. Press coverage collapsed the comparison into a single sentence, and the collapsed version, not the qualified one, is what shows up in later citations.

The second paper gets folded into the first

Five years later, Holt-Lunstad published a follow-up meta-analysis in Perspectives on Psychological Science that separated social isolation, loneliness, and living alone as distinct predictors, producing three separate odds ratios: 1.29 for isolation, 1.26 for loneliness, and 1.32 for living alone. This is a more careful and more clinically useful paper. It shows that isolation — the structural fact of having few social ties — and loneliness — the subjective experience of lacking connection — are not interchangeable, and that both predict early mortality independently even after adjusting for health status. It also notes something that rarely survives into the policy literature: the effects were stronger in samples with an average age under 65, which cuts against the common assumption that isolation is mainly an old-age problem.

In practice, the 2015 paper’s more precise, disaggregated odds ratios and the 2010 paper’s single pooled 50% figure get used almost interchangeably in secondary sources. A reader encountering “loneliness increases mortality risk by 26%” in one document and “social connection increases survival likelihood by 50%” in another is not looking at two independent findings that happen to be close. These are two different pooled statistics, from two different meta-analyses, measuring overlapping but not identical constructs, and reported on different effect-size scales. Neither the National Academies’ 2020 report nor the UK’s 2018 strategy is careless about this; both cite the underlying meta-analyses correctly. The blending happens further downstream, in summaries of summaries, where the citation survives but the scale and construct do not.

Where the number lands in policy documents

By the time the figure reaches a policy document, it typically functions as a justification for taking the topic seriously at all, rather than as a number that argues for a specific intervention. The UK’s 2018 strategy uses the broader evidence base, including Holt-Lunstad’s work, to justify embedding loneliness measurement into national statistics and funding social prescribing — a genuinely new institutional commitment, but one whose case does not depend on the precise magnitude of the mortality figure. The National Academies’ 2020 report on older adults is more careful still, and its 2020 clinical commentary in the American Journal of Geriatric Psychiatry goes further, arguing for routine assessment of isolation in health care settings on the strength of the broader mortality literature rather than any single effect size. Murthy’s 2020 book uses the figure as an opening argument for treating loneliness as a public health matter, which is a reasonable rhetorical use of a genuine finding, though the book does not carry the caveats that accompany the original meta-analysis.

AARP’s 2018 survey of adults 45 and older is a useful contrast because it does not depend on the mortality figure at all. It measured loneliness directly using the 20-item UCLA Loneliness Scale in a national sample of 3,020 respondents and found that a third of respondents were lonely, with network size and diversity as the strongest predictors. That finding stands on its own without borrowing weight from a mortality statistic collected in different populations for different purposes. The contrast is instructive: some parts of the policy literature build directly on new measurement, and some borrow authority from an older, more famous number because it is famous rather than because it is the most relevant evidence for the point being made.

What would settle the underlying claim

Holt-Lunstad’s own 2021 review, in the American Journal of Lifestyle Medicine, makes the case for treating social connection as a modifiable preventive factor on par with diet, exercise, and smoking cessation. That argument depends on evidence of a different kind than the mortality meta-analyses supply: intervention trials showing that increasing connection changes downstream health outcomes, not just observational studies showing that more-connected people already have better outcomes. That evidence base is thinner. Almost everything currently cited, including the 50% figure itself, is observational. A study capable of settling the causal question would need to randomize an intervention that changes social connection — not merely measure it — and follow mortality or a validated health proxy over years, with isolation and loneliness measured and reported separately rather than folded into a single construct. Until that exists, the honest version of the 50% figure is: a well-replicated association, at a scale worth taking seriously, that has not yet been shown to reverse when acted upon. That is a smaller claim than the one now printed on cabinet office letterhead, but it is the claim the data actually support.