What Time-Diary Data Adds That Surveys Cannot

Ask an American adult whether they feel lonely, and the answer varies by which survey happens to be asking. Ask what they did with their time yesterday, and the record does not depend on their mood that day. These are two different kinds of measurement, and they have been drifting apart in ways worth examining directly rather than folding into a single “loneliness epidemic” narrative.

Two instruments, two kinds of fact

Self-report loneliness scales, most commonly the UCLA Loneliness Scale, ask a person to rate statements like “I feel left out” or “I lack companionship.” The result is a subjective state: how connected a person feels, regardless of how much actual contact occurred. The Harvard Graduate School of Education’s Making Caring Common project found 36% of American adults reporting serious loneliness in early 2021, including 61% of adults aged 18 to 25. Cigna’s 2020 survey, using a different instrument and a different sampling frame, put overall loneliness at 61%, with 73% of workers aged 18 to 22 affected. These are not the same question asked twice. They are different questions, calibrated differently, producing numbers that cannot be stacked on the same chart without a footnote explaining why they disagree by 25 points.

The American Time Use Survey, run by the federal government, does not ask how anyone feels. It asks respondents to reconstruct their previous day in detailed intervals and codes what they were doing and, critically, with whom. It is an administrative-adjacent measure: not a feeling, but a count of face-to-face social time, hours per week, aggregated across a large and demographically representative sample. Reporting on that data, covered by WBUR in February 2024, found that face-to-face socializing has been declining in the United States, a decline that predates the pandemic and continued after it. That is a behavioral fact, not a felt one.

Where the two measures should meet, and where they diverge

If loneliness were simply the subjective readout of an objective decline in social contact, the two measures should move in lockstep: less time with others, more people reporting the feeling of being alone. Some of that correlation shows up. The AARP Foundation’s 2018 survey of adults 45 and older, using the same UCLA scale as the academic literature, found the strongest predictors of loneliness were the size and diversity of a person’s social network and physical isolation — structural facts, not just mood. Among respondents who reported speaking with neighbors, 33% were lonely; among those who never had, 61% were. That is a fifteen-point-plus gap tied to an observable behavior, which is exactly the kind of relationship a time-use record should also be able to detect independently.

But the two measures do not move at the same rate, and they do not always move in the same direction across groups. The Survey Center on American Life at AEI found that the share of Americans reporting no close friends rose from 3% in 1990 to 12% in 2021, with the collapse concentrated among men — 55% of men reported six or more close friends in 1990, versus 27% in 2021, and men reporting zero close friends rose fivefold. That is closer to a structural, countable claim than a mood-based one, since it asks about the existence of relationships rather than a feeling. It aligns with time-use data showing declining face-to-face contact. Where the picture gets harder to reconcile is with feeling-based prevalence figures, which have not shown a comparably smooth decades-long trend; they spike around discrete events, particularly the pandemic, in a way that raw hours-of-contact data does not fully explain.

Gallup’s 2023 global survey offers a further complication. Surveying 142 countries, it found 24% of people worldwide report feeling very or fairly lonely, but the demographic pattern cuts against a simple contact-hours story: adults aged 19 to 29 report the highest loneliness, at 27%, while adults 65 and older report the lowest, at 17%. Older adults are, on most behavioral measures, less socially active than young adults — they have smaller networks, more time alone, higher rates of living alone. If loneliness simply tracked observed social contact, older adults should report more of it, not less. They do not. Whatever is driving the self-reported figure among young adults is not simply a deficit in hours spent with other people.

The federal surveillance case

The CDC’s Morbidity and Mortality Weekly Report, published in June 2024, illustrates a separate comparability problem: the date on the report is not the date of the underlying behavior. The MMWR data describe conditions among U.S. adults in 2022, and it is worth being precise that this is a two-year-old snapshot dressed in a current publication date. The report links loneliness and lack of social and emotional support to heart disease, stroke, dementia, type 2 diabetes, depression, anxiety, and premature mortality, all self-reported associations drawn from surveillance instruments, not linked to individual medical records or independently verified contact logs. This is standard practice for federal surveillance and not a flaw specific to the CDC, but it means the “2024 loneliness figures” that circulate in policy discussion this year are frequently 2022 self-reports being cited as though contemporaneous.

The Surgeon General’s 2023 advisory, which drew on this same broad self-report literature, estimated that approximately half of U.S. adults experience loneliness and drew the widely repeated comparison to mortality risk from smoking up to 15 cigarettes daily. That comparison is drawn from the earlier meta-analytic literature on social isolation and mortality, not from any single time-use or administrative dataset, and it is a claim about relative risk association, not about the prevalence figure itself. The advisory is candid that it is synthesizing self-report survey data across sources of varying vintage and instrument, which is a reasonable thing to do in a policy document but a different exercise from measuring behavior directly.

Why the mismatch is not just noise

A 2023 review in BMC Public Health mapping the state of loneliness and social isolation research flagged inconsistent measurement as a structural barrier to comparing findings across studies. That is the diplomatic version of a blunter point: self-report loneliness and observed social contact are not two readings of the same underlying variable with different amounts of error. They are two different variables. Isolation is a property of a network — how many people, how often, how diverse. Loneliness is a subjective gap between the connection a person has and the connection they want. A person with a small, unchanging network of two close relationships they value can score low on loneliness scales and would show as isolated on any objective contact-hours measure. A person with dozens of weekly interactions who feels unseen in all of them can score high on loneliness while showing as highly connected on the time-use record. Both patterns exist in the literature, which is why researchers now treat isolation and loneliness as independently predictive of poor health outcomes rather than as one thing measured two ways.

The AARP’s neighbor-contact finding and the AEI friendship data point toward one honest reconciliation: the behavioral decline is real and probably a partial driver of rising loneliness, but the two measures should never be treated as interchangeable or substitutable in a single argument. A time-use survey can tell a policymaker that face-to-face contact hours have fallen. It cannot tell them how many of those remaining hours felt meaningful, and a loneliness scale cannot tell them how many hours of contact preceded the survey.

What would close the gap

The strongest test of the relationship between the two measures would be a longitudinal study that collects time-diary data and a validated loneliness scale from the same individuals at the same intervals, over years, rather than pooling separate cross-sectional surveys of different populations run by different organizations on different schedules. No source reviewed here does this. Until one does, the honest position is that self-report and behavioral measures of social disconnection are correlated in aggregate, diverge sharply by age and gender, and should be reported side by side rather than merged into a single number.