The One-in-Four Figure Rests on a Telephone and Mail Survey Frame

The claim that roughly one in four adults aged 65 and older is socially isolated appears, in some form, in nearly every policy document on ageing and connection published since 2020. It came from the National Academies of Sciences, Engineering, and Medicine’s 2020 consensus report, and it has since been repeated by the Surgeon General’s 2023 advisory and cited in health-system planning documents on the strength of that repetition. Repetition, however, is not replication. The figure deserves a closer look at where it came from and what population it actually describes.

What the National Academies report drew on

The National Academies report was a consensus report, not a primary survey. Its authors synthesized existing prevalence estimates rather than fielding new data collection, and the one-in-four figure for social isolation among older adults reflects that synthesis rather than a single freshly designed study with a stated sampling frame the report generated itself. This is a routine and legitimate way to produce a consensus estimate. It is also a way that a number can travel considerably further than its original sampling design was built to support, because each downstream citation tends to drop the caveats that accompanied the original synthesis.

The clinician-facing commentary on the report, published later that year in the American Journal of Geriatric Psychiatry, is useful here precisely because it was written for an audience that would actually try to act on the number — clinicians deciding whether and how to screen for isolation in practice. That commentary argues for routine assessment in clinical settings, which is a sensible recommendation, but it also implicitly concedes that “routine assessment” is necessary because the population-level estimate cannot substitute for individual measurement. If the underlying prevalence figure were sampling-robust in the way a well-specified probability survey is, the case for universal screening would rest more on efficiency than on uncertainty about who is actually isolated.

The recruitment problem specific to older populations

Prevalence estimates for social isolation and loneliness among older adults face a structural difficulty that estimates for younger cohorts do not: the people least reachable by a given survey mode are disproportionately the people the survey is trying to count.

A national survey conducted by phone will systematically undersample older adults who have no landline, who screen calls, who have hearing loss that makes phone interviews difficult, or who live in institutional settings such as skilled nursing facilities. A survey conducted by mail will undersample those with vision impairment, cognitive decline, or no fixed address. A web panel will undersample the substantial share of adults over 75 who do not use the internet regularly. Each of these excluded groups is plausibly more isolated, not less, than the adults who remain reachable and willing to respond. If isolation correlates with exactly the characteristics that predict non-response — reduced mobility, sensory impairment, thinner social networks to help with survey completion — then any single-mode survey frame will tend to understate true prevalence, and by an unknown and unmeasured amount.

This is not a hypothetical concern specific to one survey. The AARP Foundation’s 2018 national survey of 3,020 adults aged 45 and older is a comparatively strong instrument on other grounds — it used the 20-item UCLA Loneliness Scale rather than a bespoke question, which makes it more comparable to the academic literature than many government surveys are. But it still depended on respondents being reachable and willing to complete a lengthy interview, and the AARP report itself identifies physical isolation as one of the top predictors of loneliness. A predictor of loneliness is, almost by definition, also a predictor of being hard to survey.

What the CDC’s 2024 surveillance data adds, and does not

The CDC’s Morbidity and Mortality Weekly Report on loneliness, published in June 2024, offers federal surveillance estimates of loneliness and lack of social and emotional support among U.S. adults. It is worth being precise about dates here: the report was published in 2024, but the underlying survey data is from 2022. That two-year gap matters less for a slow-moving demographic pattern than it would for, say, employment, but it is a gap the report itself must be read against.

The CDC estimate is a genuine improvement over relying solely on the 2020 consensus figure, because it is drawn from a large federal surveillance system with documented methodology rather than a synthesized secondary estimate. But it measures loneliness and lack of social and emotional support, not social isolation as the National Academies defined it. Isolation is a structural property of a person’s network — how many contacts they have, how often they see them, whether they live alone. Loneliness is the subjective experience of finding that network insufficient. A person can be objectively isolated by any structural measure and report no loneliness at all, and the reverse is well documented too. Citing the CDC’s loneliness figure as though it confirms the National Academies’ isolation figure treats two different constructs as interchangeable, which the survey instruments themselves do not support.

The 2023 BMC Public Health review of the state of loneliness and social isolation research makes a version of this point at the level of the whole field: inconsistent measurement across studies is a structural barrier to comparing findings, not an incidental one. A reader trying to track whether isolation among older adults has risen, fallen, or stayed flat since 2020 cannot actually answer that question by comparing the National Academies figure to the CDC figure, because they were not measuring the same thing, using the same instrument, on the same sampling frame, at the same time.

What this means for policy built on the one-in-four figure

None of this means the underlying concern is overstated. The Surgeon General’s 2023 advisory places social disconnection’s mortality risk alongside smoking up to fifteen cigarettes a day, a comparison drawn from the mortality meta-analyses rather than from the isolation prevalence figure itself, and that comparison does not depend on getting the one-in-four number exactly right. The direction of the evidence — that isolation and loneliness carry real health costs for older adults — is not in serious dispute.

What is in dispute, or ought to be, is any policy document that treats one-in-four as a precise population parameter suitable for, say, projecting the number of Medicare beneficiaries who would benefit from a specific screening program, or for measuring whether an intervention has moved the needle. A figure built by synthesizing prior estimates, several of which likely undersampled the hardest-to-reach and plausibly most isolated older adults, is better treated as a floor than as a point estimate. The true prevalence could reasonably be higher; there is no equivalent structural reason to think it is lower.

A stronger estimate would need a sampling frame explicitly designed to reach institutionalized older adults, those without stable phone or internet access, and those with sensory or cognitive impairment that standard survey modes screen out by default — not as an afterthought correction applied to a telephone sample, but as a population the design targets from the start. Until such a frame exists, the one-in-four figure should travel with its provenance attached, not as a bare statistic.