One Construct, Four Instruments: Why National Loneliness Figures Do Not Compare

Four figures, all published within roughly five years of one another, all describing loneliness prevalence in general adult populations: 24% worldwide, approximately 50% of US adults, 61% of US adults, one in three US adults aged 45 and over. They are not in conflict. They are answers to different questions.

Gallup’s 142-country survey, published in October 2023, found 24% of people reporting they felt very or fairly lonely. The US Surgeon General’s advisory, published in May 2023, put the figure at approximately half of US adults. Cigna’s 2020 workplace report found 61% of US adults reporting they sometimes or always feel lonely, up seven points from the prior year. The AARP Foundation’s 2018 survey of 3,020 midlife and older adults found one in three lonely.

The instruments behind those four numbers differ in the number of items, the response format, the threshold for counting as lonely, and in one case whether the word “lonely” appears at all. Comparing them as though they measure the same thing produces conclusions that the data cannot support — most commonly, the claim that loneliness is rising, or that one country is lonelier than another.

The direct question and its ceiling

Gallup’s approach, applied in the survey conducted with Meta between June 2022 and February 2023 across roughly 1,000 respondents in each of 142 countries, asks directly. Respondents report how lonely they feel; 24% answered very or fairly lonely, and 49% reported not feeling lonely at all.

The direct question is the cheapest instrument to field and the easiest to translate, which is why it dominates cross-national work. Its weakness is well documented in the measurement literature: asking people whether they are lonely invites social desirability bias, and the size of that bias plausibly varies by culture. If admitting loneliness carries more stigma in one country than another, the direct question will underestimate prevalence unevenly across the comparison — precisely the comparison the instrument is being used to make.

Gallup’s own analysis, published in November 2023, makes a related point that is easy to miss. Across the 142 countries, 72% of respondents felt very or fairly socially connected — roughly 3.2 billion people in the surveyed populations, which exclude China and therefore cover about 77% of the world’s adults. Connectedness and loneliness are not simple inverses. A country can rank well on one and poorly on the other. That non-inversion is itself evidence that the two questions are tapping distinguishable states, and a reason to be wary of treating any single item as a summary of a population’s social health.

The UCLA scale and why AARP’s number is the most portable

The AARP Foundation’s 2018 survey used the 20-item UCLA Loneliness Scale rather than a bespoke instrument. This matters more than the finding it produced. The UCLA scale never uses the word “lonely”; it asks about frequency of experiences such as feeling left out or feeling that one’s relationships are superficial, and derives a score. That design substantially reduces the stigma problem, and because the scale has been used across decades of academic work, a figure derived from it can be placed against the wider literature.

One in three adults aged 45 and over, from 3,020 respondents. The survey also found that 33% of those who had spoken to their neighbours were lonely, against 61% of those who never had, and identified network size and diversity alongside physical isolation as the strongest predictors.

Note what that comparison does and does not establish. The neighbour finding is cross-sectional. Speaking to neighbours may reduce loneliness, loneliness may reduce the inclination to speak to neighbours, or a third factor — mobility, tenure, neighbourhood density — may drive both. The 28-point gap is real; its direction is not established by this design.

The UK’s institutional answer

The UK’s 2018 strategy, the first national loneliness strategy published by any government, did something structurally different from any of the surveys above. Rather than commissioning a prevalence estimate, it embedded loneliness measurement into the Office for National Statistics and funded social prescribing.

Embedding the question in a national statistical agency changes what the data can do. A one-off survey produces a snapshot; a standing measure in a recurring official survey produces a time series with consistent sampling, consistent wording, and consistent fieldwork procedures. The trade-off is that the resulting series is comparable to itself and to very little else. A figure produced by ONS wording cannot be set against a figure produced by Cigna’s wording and read as a difference between the UK and the US.

Japan took the ministerial route rather than the statistical one, appointing Tetsushi Sakamoto to a newly created cabinet post in February 2021 — the second country after the UK to do so. The two governments held the first bilateral ministerial meeting dedicated to loneliness policy in June 2021, which is where the measurement problem became a diplomatic one: two states committing to a shared policy area without a shared metric.

Japanese researchers, meanwhile, produced something the survey literature mostly lacks. The JACSIS study, published in 2023 with national data covering 2020 to 2021, is longitudinal rather than cross-sectional. It can therefore speak to change over the pandemic period rather than to a single snapshot. That distinction is the whole difference between “prevalence was higher in 2021 than 2020” and “loneliness increased” — and only the second requires following the same people.

What the US figures actually show about change

The three US figures — 61%, approximately 50%, 36% — are frequently arranged into a narrative about rising loneliness. They do not support one.

Cigna’s 61% counts anyone answering “sometimes or always”. Harvard’s Making Caring Common report, published in February 2021, found 36% reporting serious loneliness, a deliberately higher threshold, alongside 61% of young adults aged 18 to 25 and 51% of mothers with young children. The Surgeon General’s approximately half is a synthesis figure drawn across sources rather than a single field result. Three different thresholds applied to three different samples in three different years produce three numbers whose ordering tells you about the thresholds, not about the trend.

There is one genuine within-instrument comparison in that set: Cigna’s seven-point year-over-year increase, same wording, same design. That is a defensible statement about change in what Cigna measured between two fieldwork periods.

Publication dates compound the problem. The Surgeon General’s advisory appeared in May 2023 but synthesises studies whose fieldwork runs years earlier. Gallup’s October 2023 publication reports data collected between June 2022 and February 2023. The BMC Public Health review of the field, published in June 2023, names inconsistent measurement as a barrier to comparing findings across studies; the lag between fieldwork and publication is a second barrier that receives less attention and produces confident claims about “current” loneliness levels drawn from surveys two or three years old.

The construct that gets lost

Every figure discussed so far measures loneliness — a subjective state. None of them measures social isolation, which is a structural property of a person’s network: how many contacts, how frequent, how varied.

The National Academies estimated in 2020 that roughly one quarter of US adults aged 65 and over are socially isolated. That number is not comparable to any loneliness figure, and the two constructs move independently. 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 — three related but separable exposures, each independently predictive of early mortality.

An isolated person may not be lonely; a person with a dense network may be. When a government reports a loneliness figure and a health system reports an isolation figure and a journalist combines them, the resulting composite describes nothing. The 2024 review of the epidemic framing in PMC returns to this point: the strength of the case for treating disconnection as a population health issue does not depend on any single prevalence number, and is weakened when incompatible numbers are stacked.

What a comparable estimate would require

The WHO Commission on Social Connection, established in November 2023 and co-chaired by Vivek Murthy and Chido Mpemba, is the first body positioned to standardise this. The most useful thing it could produce is not another prevalence estimate but a minimum measurement set: one validated multi-item loneliness scale with published translations and tested cross-cultural invariance, one structural isolation measure with defined thresholds, both fielded on the same respondents, at fixed intervals, with a longitudinal panel component in at least a subset of countries.

Until something like that exists, the honest form of a cross-national claim is narrow. Gallup’s 24% and the Surgeon General’s approximately half do not describe two populations. They describe two questions.