What the Gallup-Meta Global Connection Survey Leaves Out

When the World Health Organization launched its Commission on Social Connection in November 2023, co-chaired by Vivek Murthy and Chido Mpemba, it needed a global baseline. The figure it reached for came from Gallup and Meta’s State of Social Connections report: 72% of people worldwide feel very or fairly socially connected, and 24% report feeling lonely. These numbers now anchor a policy framing that treats social disconnection as a measurable, comparable, worldwide condition, the same way global health bodies track obesity or vaccination coverage.

The survey behind that number deserves more scrutiny than it has received, and the scrutiny worth doing is about who was reachable to be asked in the first place.

What the survey actually sampled

Gallup and Meta surveyed roughly 1,000 people aged 15 and older in each of 142 countries between June 2022 and February 2023, for a combined sample covering an estimated 3.2 billion people in the countries included. That is a serious undertaking, and Gallup is transparent about one major exclusion: China is not in the sample, which means the survey speaks for roughly 77% of the world’s adult population, not all of it. Gallup’s companion analysis, published two weeks later, breaks the results down by country and demographic group and is careful to note that connectedness and loneliness are not simple inverses of each other, which is itself a useful corrective to how the topline numbers get used.

What is less discussed is how the other 141 countries were reached. Gallup’s World Poll, the vehicle for this kind of study, uses telephone interviewing in countries with sufficient phone coverage and face-to-face interviewing elsewhere, with online panels used in some higher-income markets. That is a reasonable, well-established methodology for measuring attitudes and self-reported wellbeing at scale. It is a less obviously reasonable methodology for measuring social connection specifically, because the mode of contact is not neutral with respect to the thing being measured.

Why the mode of contact is not neutral here

A telephone survey reaches people who have a working telephone number, are willing to answer an unfamiliar caller, and are not so socially withdrawn that a stranger’s call goes unanswered or unreturned. An online panel reaches people with functioning internet access and enough discretionary engagement with digital platforms to have joined a panel at some point. Both of these are correlated with the very thing the survey is trying to measure. A person who has withdrawn from contact, who lives alone with no working phone line, or whose isolation is severe enough that they do not maintain an active number or email address, is less likely to appear in the sample at all. This is not a hypothetical concern for a niche subgroup. It describes exactly the population that loneliness research consistently identifies as highest-risk: people living alone, older adults with shrinking networks, and those with the fewest social ties to begin with.

The effect this produces is a form of survivorship bias built into the sampling frame. The 72% figure describes people connected enough, and resourced enough, to be surveyed. It cannot describe people disconnected enough to fall outside the frame’s reach, and there is no way to know from the report how large that excluded group is, because by definition it was not measured.

Gallup’s separate report on global loneliness, released the same week, shows a pattern that is at least consistent with a coverage problem: adults 65 and older report the lowest loneliness rate in the survey, 17%, against 27% among adults 19 to 29. That ordering runs against a substantial body of prior work. The National Academies’ 2020 consensus report put isolation among U.S. adults 65 and older at roughly one in four using clinical assessment tools, and Julianne Holt-Lunstad’s 2015 meta-analysis found social deficits were, if anything, more predictive of mortality in samples averaging under 65 than in older cohorts — a different finding, but one that does not support treating older adults as the least-lonely group in any absolute sense. There are real cultural and cohort explanations for why older respondents might report less loneliness on a self-report scale. But a sampling frame that under-reaches the most isolated older adults, those without a working phone or the mobility to answer a door-to-door interviewer, would produce the same pattern by construction, regardless of the true rate.

The instrument problem compounds the sampling problem

Separate from who was reached is what they were asked. Gallup’s single-item self-report of feeling “very,” “fairly,” “not very,” or “not at all” lonely or connected is a coarser instrument than the 20-item UCLA Loneliness Scale used in, for instance, the AARP Foundation’s 2018 survey of U.S. adults 45 and older, which found one in three reporting loneliness. Comparing Gallup’s 24% global figure to any UCLA-scale-derived national figure is comparing different instruments as much as different populations, and the BMC Public Health review of the loneliness research field published in 2023 names inconsistent measurement across studies as one of the central obstacles to building a coherent evidence base. A single-item measure is not wrong to use — it is fast, translatable across 142 countries, and cheap enough to make a global survey feasible at all — but its simplicity is a trade against precision, and that trade is rarely stated plainly when the 24% figure gets quoted.

Why this matters for what the WHO Commission does next

None of this means the Gallup-Meta estimate is worthless, and it does not mean the broader claim — that loneliness is a global, not merely high-income, phenomenon — is wrong. The U.S. Surgeon General’s 2023 advisory reached a domestically similar conclusion, that about half of U.S. adults report loneliness, using different survey instruments entirely, which is some independent corroboration that the general magnitude is not an artifact of one methodology. The concern here is narrower and more specific to policy design: a commission built to reduce global disconnection needs its baseline data to be most accurate exactly where the problem is worst, and the sampling frame used to produce that baseline has structural reasons to be least accurate there. A survey that systematically undercounts the most isolated people will understate the scale of the problem it is trying to describe, and will do so quietly, without flagging where its blind spots fall geographically or demographically.

This is a familiar shape of error in social science, sometimes called the “hard to reach” problem, and it has a technology dimension that is easy to overlook when the funder of the survey is a technology company with an obvious interest in the framing of digital connection. Phone and internet-based reach is not a neutral instrument for measuring the human condition it is being used to describe; it is itself a proxy for connectivity, which is adjacent to but not identical with social connection.

What would strengthen this evidence is not a bigger sample but a different one: linkage to administrative or registry-based data — postal records, health system rolls, or census enumeration — that does not depend on a person being reachable by phone or panel to be counted, paired with in-person follow-up in a representative subsample of the hardest-to-reach group. That is a harder and more expensive study to run than a global phone poll. It is also the only way to find out whether the 24% figure is a genuine global rate or a floor.