What Time-Use Diaries Show That Loneliness Surveys Do Not

Two kinds of evidence claim to describe the same phenomenon and produce different timelines. Self-report surveys ask people how lonely they feel, usually on a validated scale, and those numbers have moved sharply in the past decade. The Cigna Group’s 2025 survey found 57% of American adults report loneliness, based on fieldwork from May and June 2024. Harvard’s Making Caring Common project found 36% of Americans reporting serious loneliness in early 2021, with 61% among adults 18-25. AARP’s 2025 follow-up to its 2018 survey of adults 45 and older reports rising loneliness in that population, using the same 20-item UCLA Loneliness Scale as the earlier wave, which makes the comparison unusually clean for this literature.

Set against this is a different kind of data: what people actually did with their time, recorded not by asking them to rate a feeling but by asking them to reconstruct their day. The American Time Use Survey, a federal diary-based instrument, shows a decline in face-to-face socializing that predates the pandemic and continued afterward, as reporting on the data has documented. That decline is real and worth taking seriously. But it is gradual, and it does not track the survey-based loneliness numbers year for year. The two data sources are not disagreeing about facts so much as measuring different objects.

Two different questions

A loneliness survey asks a subjective question: does this person feel isolated, unsupported, or emotionally unseen. A time-use diary asks a structural question: how many minutes did this person spend in the physical presence of other people, doing what. These are not proxies for each other. The distinction between isolation, a structural property of a person’s network and time allocation, and loneliness, the subjective experience of deficient connection, is one of the more consistent findings in the literature, and one of the most consistently ignored in general reporting. A 2024 study in Scientific Reports examining the pandemic period found that the relationship between the two varies by age: older adults can be highly isolated by any structural measure while reporting comparatively low loneliness, and the reverse pattern shows up among people with dense but unsatisfying social contact. A 2023 review in BMC Public Health lists inconsistent measurement as one of the central barriers to comparing findings across the loneliness literature, and this is a clean example of why.

Time-use diaries have a specific advantage: they are not asking someone to evaluate their own emotional state, so they are not vulnerable to the biases that affect self-report, such as social desirability, current mood at the moment of the survey, or the framing of the question. Gallup’s 2023 global survey found 24% of people worldwide report feeling lonely, using a direct self-report item; a survey asking about isolation, or about minutes spent alone, would very likely produce a different number from the same population, because it would be measuring something else. That is an instrument mismatch, not an inconsistency to be explained away.

Diary methods have their own limitations. They rely on recall accuracy over a single day or a small number of days, they are expensive to run at scale, and they say nothing about the quality of the contact recorded. Two hours spent in the same room as family members watching television is coded the same way as two hours of engaged conversation. A structural measure this coarse cannot capture what several qualitative reviews of social prescribing programs have found: that the benefit participants describe often comes from purposeful, structured engagement rather than proximity alone. Time-use data is a floor, not a full account.

The administrative layer: what closed

A third kind of evidence sits underneath both of these: records of what physical infrastructure exists for people to gather in. A 2025 study in Health & Place tracked twelve categories of what Ray Oldenburg called third places — coffee shops, libraries, museums, recreation centers, restaurants — and found closures across every category between 2019 and 2021, concentrated in census tracts with higher social vulnerability and in rural areas. The University of Colorado Boulder’s summary of that research frames the loss of senior centers and churches as a driver of loneliness rather than a downstream symptom of it.

This is a genuinely different data type again: administrative and commercial records of what exists, independent of anyone’s self-report and independent of how any individual spent their time. It answers neither of the first two questions directly. But it constrains the plausible range of both. If third-place density fell in a given county, then the opportunity for the kind of contact time-use diaries would record has fallen too, whether or not any individual in that county reports feeling lonelier. The three layers — subjective report, behavioral record, structural inventory — describe a single situation from three angles that do not have to move together, and in this literature they frequently do not.

Where technology complicates the comparison

None of the three measures above was designed with an eye to how time online should be classified, and this is becoming a live problem. A commentary from George Mason University’s public health faculty this year frames AI companionship tools as a substitution risk rather than a straightforward remedy for loneliness: a person can report feeling less lonely while spending less time in the kind of in-person contact a time-use diary would capture as face-to-face socializing. If that substitution is happening at scale, self-report loneliness scores and behavioral contact time could genuinely diverge, not because one instrument is flawed, but because a new category of interaction does not map cleanly onto either one. Time-use surveys generally code time on a device without distinguishing a video call with a grandchild from scrolling alone from a conversation with a chatbot. Loneliness scales ask about felt connection without asking what supplied it.

What would settle more

A study designed to close this gap would need three things current work rarely combines: a validated loneliness scale, a time-use diary or passively logged contact data from the same respondents in the same window, and a category system granular enough to separate human-mediated digital contact from AI-mediated contact from time alone. Almost nothing in the current literature does all three. Until it does, the honest description of the evidence is that self-reported loneliness has risen faster and more visibly than behavioral measures of face-to-face contact, that structural third-place infrastructure has genuinely eroded in specific and unevenly distributed ways, and that no dataset yet available can say with confidence how much of the gap between the subjective and behavioral numbers is measurement artifact and how much is a real change in what contact means to the people reporting it.