What the Loneliness Surveys Cannot Tell Policymakers

Four surveys published between 2018 and 2021 put the share of lonely American or British adults somewhere between 22% and 61%. That is not a narrow band of measurement error. It is a fourfold range, and it means that “how much loneliness is there” — the question policymakers most often ask of this literature — cannot currently be answered with one number.

The Cigna 2020 report found 61% of U.S. adults sometimes or always lonely, up seven points from the year before. The Harvard Making Caring Common survey, fielded roughly a year later, found 36% of Americans reporting serious loneliness. The AARP Foundation’s 2018 survey of adults 45 and older found one in three lonely. None of these figures is wrong. They are answers to different questions, asked of different populations, using different instruments, at different points in a fast-moving two-year period that included a pandemic. Treating them as interchangeable measurements of a single underlying quantity is the most common analytical error in loneliness reporting, and it is the one policy briefs need to stop making first.

Why the numbers diverge

The AARP survey used the 20-item UCLA Loneliness Scale, the closest thing the field has to a standard instrument, and one with decades of use in academic research. That gives its one-in-three figure real comparability to prior studies using the same tool. Cigna’s index and the Harvard survey used their own constructed measures, asking about frequency of feeling lonely or isolated in language that does not map cleanly onto the UCLA scale’s items about feeling “left out” or lacking companionship. A single yes/no or sometimes/always question about loneliness will almost always produce a higher prevalence estimate than a validated multi-item scale, because it invites a lower threshold for endorsement and offers no way to distinguish transient low mood from a sustained state.

Age composition compounds the problem. AARP surveyed only adults 45 and older, a population the National Academies’ 2020 consensus report estimates to be roughly one-quarter isolated (a distinct construct from loneliness, discussed below) rather than one-third lonely. Harvard’s survey, by contrast, oversampled younger adults and reported 61% loneliness among 18-to-25-year-olds specifically. Any single national figure that averages across age groups with such different baseline rates will move substantially depending on how the sample is weighted, and most published summaries do not show that weighting.

Timing matters as much as instrument. The Cigna and Harvard figures were both collected during the pandemic; the AARP and UK government figures predate it. A researcher stacking these four numbers into a single trend line would be describing something closer to noise from instrument-switching than a real trajectory of national loneliness.

Isolation, loneliness, and the network questions no survey answers well

Separately from the measurement inconsistency, there is a conceptual gap the survey data does not close. Isolation — the objective size and density of a person’s network — and loneliness — the subjective feeling of lacking connection — are correlated but not identical, and the policy implications of each are different. The National Academies report estimates about one-quarter of adults 65 and older are socially isolated. That is a network-structure claim: few contacts, infrequent contact, few group memberships. It says nothing directly about whether those people feel lonely, and Holt-Lunstad’s 2015 meta-analysis found loneliness (OR = 1.26), isolation (OR = 1.29), and living alone (OR = 1.32) each independently predicted early mortality, meaning none of them is a proxy for the others.

Most national surveys ask about loneliness — the feeling — because it is easier to elicit in a single self-report item. Isolation requires asking about network size, contact frequency, and group membership, which the American Enterprise Institute’s 2021 friendship survey does, finding 12% of Americans report no close friends at all, up from 3% in 1990, and a fivefold increase in men reporting zero close friends since that year. That is a genuine structural finding, and a troubling one, but it is not the same measurement as the Cigna or Harvard loneliness figures, and none of these surveys was designed to let a reader connect a specific person’s isolation score to their loneliness score. Cross-sectional national surveys, by design, capture a population snapshot, not the pathway by which shrinking networks turn into felt loneliness for some people and not others.

What this means for anyone trying to act on the numbers

A government or health system reading these reports for guidance faces two practical problems the topline prevalence figures do not solve.

The first is targeting. The UK’s 2018 loneliness strategy set out to embed loneliness measurement into the Office for National Statistics precisely because it needed a consistent, repeatable national instrument rather than a patchwork of one-off surveys — an implicit acknowledgment that the pre-2018 evidence base could not tell government which populations to prioritize with any confidence. Without a standardized instrument applied longitudinally to the same population, a health system cannot distinguish a genuine rise in loneliness from a change in how the question was asked, and cannot tell whether an intervention lowered loneliness or simply coincided with a survey redesign.

The second is mechanism. None of these four surveys is longitudinal, and none was designed to test whether an intervention changes loneliness or isolation. They describe prevalence at a moment in time, cross-sectionally, which is a different scientific task from establishing what reduces the state they describe. A survey finding that 61% of employed Gen Z workers report loneliness (Cigna’s figure) is useful for signalling scale to an employer or a legislature. It does not tell that employer which workplace policy, if any, would lower that number, because the survey was never structured to test one.

What a better instrument would look like

The gap in this literature is not more prevalence surveys — there are already several, asking different questions of different populations at different times, in a way that makes cross-study comparison close to meaningless. What would strengthen the evidence is a single validated instrument, ideally the UCLA scale or a directly calibrated equivalent, applied to the same panel of respondents at regular intervals, alongside a separate and equally consistent measure of network structure so that isolation and loneliness can be tracked as the two distinct variables they are. Layered onto that, intervention studies — ideally randomized rather than pre/post — that measure whether a specific program changes either variable, rather than surveys that simply document how many people report feeling lonely this year compared to last.

Until that exists, the honest answer to “how much loneliness is there, and is it rising” is that the current data cannot support a single confident figure, and any policy brief that quotes one number without naming its instrument and its year is quietly discarding the very uncertainty a reader needs in order to judge what the number is worth.