What 'Half of Americans Are Lonely' Actually Measured

The U.S. Surgeon General’s May 2023 advisory, Our Epidemic of Loneliness and Isolation, states that approximately half of American adults report experiencing loneliness. That figure has since been repeated in congressional testimony, agency strategy documents, and news coverage as though it were a single, stable measurement of the country’s emotional state. It is not. It is a synthesis drawn across several surveys that used different instruments, different time frames, and different populations, and that produced meaningfully different numbers when read on their own.

This matters for a policy brief because “half of Americans are lonely” implies a target population and a baseline against which an intervention’s success could later be judged. Neither is well defined by the advisory’s summary statistic, and the sources beneath it disagree with each other more than the headline figure suggests.

Where the number comes from

The advisory does not report a single new survey. It synthesizes prior work, and the studies it draws on vary widely. Cigna’s 2020 workplace loneliness survey found 61% of U.S. adults reporting they sometimes or always feel lonely, a figure that had risen 7 points from the prior year. Harvard’s Making Caring Common project, surveying in 2020 and publishing in February 2021, found 36% of Americans reporting serious loneliness — using a stricter threshold than “sometimes” — but 61% among adults aged 18 to 25 specifically. AARP’s 2018 survey of adults 45 and older, using the 20-item UCLA Loneliness Scale, found one in three lonely. These are not the same question asked three times. They are three different questions, asked of three different age ranges, at three different points before, during, and after the acute phase of the pandemic, each yielding a number in a different band.

“Half” is a reasonable midpoint gloss across a set of measurements ranging from roughly 33% to 61%. It is not a measurement in its own right. Anyone using it to argue that loneliness prevalence has a stable, comparable value across time is asking the number to do work the underlying data cannot support.

Instrument mismatch, not just noise

The differences are not simply sampling variation around a true rate. They are largely explained by what each survey asked. Cigna’s instrument asked about frequency of feeling lonely in binary terms — sometimes or always versus rarely or never — which tends to inflate prevalence relative to instruments requiring a stronger endorsement. The UCLA Loneliness Scale used by AARP asks about specific experiences (feeling left out, feeling isolated from others, lacking companionship) and aggregates them into a score, which is closer to the instrument used across most of the academic mortality literature, including Julianne Holt-Lunstad’s 2015 meta-analysis in Perspectives on Psychological Science. Harvard’s “serious loneliness” threshold was defined more narrowly still.

A 2023 review in BMC Public Health on the state of loneliness and social isolation research flags exactly this problem at the field level: inconsistent measurement is one of the primary barriers to comparing findings across studies, let alone across countries. When a policy document averages across instruments of different strictness, the resulting number does not describe a population; it describes an artifact of which surveys happened to be available to cite.

Gallup’s 2023 global survey, conducted across 142 countries and using yet another framing — asking respondents directly whether they felt lonely “a lot of the day yesterday” or in similarly bounded terms — found 24% of people worldwide reporting loneliness, with young adults aged 19–29 highest at 27% and adults 65 and older lowest at 17%. That is a fundamentally different prevalence estimate from anything in the U.S.-only literature, and it inverts the popular assumption that loneliness is chiefly an older-adult problem. The Gallup figure is lower than nearly every U.S. domestic estimate, which likely reflects both instrument and population differences, not a genuine gap in loneliness between the United States and the rest of the world.

What the advisory does establish

None of this means the Surgeon General’s advisory is wrong to treat loneliness as a public health concern. Its strongest evidentiary claim is not the half-of-adults prevalence figure but the comparison to mortality risk: it states that the mortality risk associated with social disconnection is comparable to that of smoking up to fifteen cigarettes a day. That claim rests on the meta-analytic literature — principally Holt-Lunstad’s earlier work — which used consistent effect-size measures (odds ratios for isolation, loneliness, and living alone) across pooled studies rather than a single cross-sectional prevalence survey. That comparison is on firmer ground than the prevalence headline, because it is built from effect sizes aggregated across many measurement approaches rather than a single self-report percentage presented as a population fact.

The advisory’s real policy contribution is not a prevalence number at all. It is the argument, laid out in its six-pillar National Strategy to Advance Social Connection, that federal agencies should treat social connection as an infrastructure and health-system issue rather than a private matter. That argument does not depend on whether the true figure is 33%, 50%, or 61%.

What a policy document should do instead of averaging

A brief that wants to state prevalence honestly has three better options than a single national percentage.

First, name the instrument and the year alongside any figure. “36% of U.S. adults reported serious loneliness in a Harvard survey fielded in 2020” is a defensible sentence. “Half of Americans are lonely” is not, because it conceals which survey, which year, and which threshold produced it.

Second, separate age groups rather than pooling them. The consistent finding across the Harvard, AARP, and Gallup surveys is that young adults report loneliness at rates well above the population average — 61% among 18-to-25-year-olds in the Harvard survey, 27% among 19-to-29-year-olds in Gallup’s global data — while adults 65 and older report it at rates below the average in the same surveys. A single blended figure erases the one distributional pattern that is actually stable across instruments.

Third, track a single instrument over time rather than switching sources year to year. The UK’s approach, embedding a consistent loneliness measure into Office for National Statistics data collection since its 2018 national strategy, is closer to what would allow a real trend line. The U.S. literature, by contrast, is a set of one-off surveys from different organizations, which makes “is loneliness rising or falling” a question the current evidence base cannot answer with confidence, whatever any single survey’s year-over-year comparison suggests within its own methodology.

The gap this leaves for evaluation

The practical stakes are evaluative. If a federal or state program is funded on the premise that it will reduce loneliness from a baseline of “half of adults,” and the eventual outcome measure uses a different instrument than whatever produced that baseline, any observed change will be uninterpretable — it could reflect the intervention, or it could reflect the switch in measurement. Given that the Surgeon General’s own summary figure is a synthesis across instruments rather than a single measured baseline, no future single survey can be validly compared against it to claim progress or decline.

A stronger evidence base would specify one instrument, applied to a nationally representative sample, repeated on a fixed schedule, with age and living-arrangement breakdowns reported by default rather than as a secondary analysis. Absent that, the honest use of the current literature is to describe a range — roughly a third to three-fifths of U.S. adults, depending on instrument and year — and to say plainly that the range, not a single percentage, is what the evidence supports.