Why Two 2025 Surveys of American Loneliness Disagree by Twenty Points

Two figures now circulate as the answer to “how many Americans are lonely.” The Cigna Group’s Loneliness in America 2025 reports 57%. An analysis of the sixth cycle of the Health Information National Trends Survey, published in the American Journal of Preventive Medicine in November 2025, reports that roughly 1 in 7 U.S. adults are severely lonely and 1 in 5 moderately lonely, for a combined moderate-to-severe figure just over 37%. Both are national, both are recent, and both describe the same country in roughly the same period. The twenty-point gap between them is not a sign that one study is wrong. It is a demonstration of how much the answer to “how lonely is America” depends on the question asked to get there.

Two different questions, not one contested fact

The HINTS-6 analysis is built on data the National Cancer Institute collected in 2022, though the paper analyzing it was published in 2025 — a publication-versus-data-date gap worth flagging on its own, since a reader citing “a 2025 study” might reasonably assume the underlying experience is more recent than it is. HINTS-6 asks about loneliness using a graded set of states — none, mild, moderate, severe — derived from a validated multi-item scale, and the published figures report the moderate and severe categories combined. That is a relatively high bar. Someone who feels lonely sometimes, but not persistently or intensely, would likely land in the “mild” category the paper does not fold into its headline number.

Cigna’s methodology sits at the opposite end. Its 57% figure, drawn from fieldwork conducted from late May to mid-June 2024 among more than 7,500 U.S. adults, appears to reflect a broader threshold — the share of people who report feeling lonely at all, rather than a subset scoring in the upper bands of a graded severity scale. This is the same pattern the U.S. Surgeon General’s 2023 advisory ran into when it settled on “approximately half” of U.S. adults as its topline: a number that sits between a low bar (any loneliness, ever) and a high bar (persistent, severe loneliness), depending on which studies fed into the estimate.

Neither approach is wrong. A public health surveillance system arguably wants the high bar — the CDC’s own 2024 surveillance report on loneliness, drawing on 2022 data, is more interested in who is at meaningfully elevated risk of the downstream harms (heart disease, dementia, depression) than in who ticked a box marked “sometimes.” A market-facing or advocacy-facing report, by contrast, has an interest in capturing the full range of people who experience loneliness at all, because that is the population a program or product might plausibly reach. Both are legitimate research questions. They are simply not the same research question, and reporting on prevalence tends to collapse them into a single number anyway.

The instrument problem is bigger than these two studies

This is not an isolated case. A 2023 review in BMC Public Health of the loneliness and social isolation literature identified inconsistent measurement as one of the central barriers to comparing findings across studies — not a footnote problem but a structural one, because prevalence estimates that use different scales, different response thresholds, and different reference periods (“right now” versus “in the past week” versus “in general”) are not measuring the same underlying construct even when they use the same word.

The UCLA Loneliness Scale, the most widely used academic instrument, typically reports loneliness on a continuous or several-tier basis rather than as a single yes/no. Surveys that instead ask a single item — “do you feel lonely?” — tend to produce different distributions than surveys using a validated multi-item scale, partly because a single item is more sensitive to mood on the day of the survey and partly because respondents interpret “lonely” itself differently depending on what else the question implies. AARP’s own longitudinal work is useful here precisely because it has kept its instrument constant: its 2025 follow-up survey of adults 45 and older, comparable to its 2018 baseline, can speak to whether loneliness in that age group has risen over seven years in a way that a one-off comparison across two different survey houses cannot. That kind of within-instrument comparison is rare in this literature, and it is exactly the design feature that lets a claim about change over time hold weight.

What each number is actually useful for

None of this means the two 2025 figures are interchangeable or that either should be discarded. The HINTS-6 moderate-to-severe estimate is the more defensible number for arguments about disease burden and health system planning, because it tracks something closer to the population at elevated risk of the mortality and morbidity outcomes catalogued in the cardiovascular and mortality literature. The Cigna figure is the more defensible number for arguments about how widespread the experience of loneliness is as a felt condition, including at levels too mild to register in a clinical framework but still meaningful to the person experiencing it, and that broader framing matters for questions about workplace culture, social infrastructure, and everyday social contact rather than for questions about disease risk specifically.

The practical error is using either figure as though it answers the other’s question. A policy brief that cites “57% of Americans are lonely” to argue for expanded geriatric care referral pathways is borrowing a number calibrated to a much broader threshold than the clinical case requires. A surveillance report that cites “37%” to argue that loneliness in America has been overstated is doing the reverse: treating a high, clinically weighted bar as though it captures the full scope of the experience.

What would resolve the disagreement

It would not take much to make these numbers comparable, and the fix is not more surveys — it is survey design discipline. A study that ran the UCLA Loneliness Scale and a Cigna-style single-item measure on the same nationally representative sample, in the same fieldwork window, would isolate how much of the twenty-point gap is instrument and how much, if any, is genuine disagreement about the underlying rate. Absent that, the responsible move for anyone citing prevalence figures is to name the instrument and the threshold every time, the same way a public health report would name the diagnostic criteria behind a disease prevalence estimate rather than reporting a bare percentage.

The broader lesson generalizes past this one comparison. Loneliness prevalence figures published in the same year, describing the same country, routinely differ by ten, twenty, even thirty points, and the difference is almost never a real disagreement about how lonely Americans are. It is a difference in what counts as lonely enough to count. Readers who want a single stable number to track over time are better served by picking one well-validated instrument, administered consistently, the way AARP has done with its 45-and-older cohort, than by averaging across studies that were never measuring the same thing to begin with.