Who Gets Left Out of the Loneliness Numbers

The National Academies’ 2020 consensus report put the figure at roughly one quarter of adults aged 65 and older living in social isolation. That number has been repeated in policy documents and news coverage for six years. It is worth asking who that quarter is drawn from, and, more pointedly, who was never eligible to be counted.

A sampling frame is the list a survey draws its respondents from — a phone exchange, a mailing address, a panel of people who agreed in advance to answer questions online. It sounds like a technical detail. For loneliness research among older adults, it is close to the whole story, because the frame determines whether the most isolated people in the population can appear in the data at all.

The instruments in circulation

AARP’s 2018 survey of adults 45 and older, and its 2025 follow-up, used the 20-item UCLA Loneliness Scale administered to a national probability sample by phone, mail, and online — a genuine strength, since the instrument matches the academic literature rather than inventing its own metric. But probability samples of this kind are built from residential address lists and working phone numbers. They reach community-dwelling adults who can be located and who can complete a lengthy self-report instrument unassisted.

The 2024 Household Pulse Survey data, analyzed in a 2026 paper on loneliness and isolation prevalence, comes from an online panel recruited by email and text message from Census Bureau address files. It is fast and cheap to field, which is why the Census Bureau uses it for rapid turnaround estimates. It is also, definitionally, a survey of people who have email or a working mobile number and are willing to complete a questionnaire on a screen.

The 2025 study using HINTS-6 data — a mailed, self-administered federal survey fielded in 2022 — requires a stable mailing address, functional literacy in the survey’s language, and the physical and cognitive capacity to fill out a paper or online form without assistance. It found that over 37% of U.S. adults experience moderate-to-severe loneliness, a figure that has circulated widely. The 2024 CDC surveillance report drawing on the same general period relies on telephone-based BRFSS methodology, which carries its own exclusion: no landline or working cell, no inclusion.

What these frames systematically drop

Every one of these methods excludes, by construction, several groups that the isolation literature identifies as high-risk: adults living in nursing homes and assisted living facilities, who are typically excluded from household-based sampling frames entirely; homebound older adults without a working phone or internet connection; people with dementia or significant cognitive impairment, who cannot reliably self-report on a lengthy scale; and non-English speakers in survey designs that do not field in their language.

This is not a minor edge case. The 2026 study on risk factors among community-dwelling socially isolated older adults makes a point of restricting itself to the community-dwelling population precisely because the institutionalized and homebound population is handled differently, or not at all, in most national instruments. A quarter isolated, drawn only from adults reachable enough to answer a survey, is a floor, not a ceiling.

Why this particular gap matters more here than elsewhere

Social isolation and loneliness are independently predictive of mortality — Julianne Holt-Lunstad’s 2015 meta-analysis put the odds ratio for isolation at 1.29 and for loneliness at 1.26, even after adjusting for health status. Isolation is a structural property of a person’s network: how many people they see, how often, how varied those ties are. That is exactly the property a restrictive sampling frame is most likely to miss, because the structural features that make someone isolated — no phone, no one to help them respond, no fixed address, cognitive decline that makes a 20-item scale impractical — are the same features that make them unreachable by a survey.

Loneliness, the subjective state, is somewhat more tractable to capture by proxy or brief item even among frail respondents. Isolation is harder, because the survey instrument itself is a social contact requiring a working channel of communication. A phone survey cannot detect someone who has no phone.

The direction of the bias

There is no serious argument that this exclusion inflates the reported rate. If anything, every plausible correction runs the other way. Nursing home residents, adults with moderate-to-severe cognitive impairment, and people who are homebound without digital or telephone access are, on prior evidence about the correlates of isolation, more likely to be isolated than the general population of community-dwelling, phone-and-internet-reachable older adults who make it into these samples. The “roughly a quarter” and “over a third” figures circulating in the literature and in policy documents are best read as estimates for a specific, more socially connected subset of the older population, not for older adults as a whole.

What would close the gap

A stronger national estimate would deliberately oversample or separately survey institutional settings using facility rosters rather than household address lists, use proxy reporting — verified against direct report where feasible — for cognitively impaired respondents, and field in-person interviews for the subset of the sample that cannot be reached by phone, mail, or web precisely because that unreachability is the variable of interest. None of the major recurring instruments cited in this literature currently do all three. Until one does, the headline prevalence figures for isolation among older adults should be read as describing the more connected end of that population, with the true rate for the full population, institutionalized and homebound residents included, almost certainly higher and currently unmeasured.