What a Single Wave of Young-Adult Loneliness Data Cannot Rule Out

Nearly every major survey published in the past five years finds the same reversal: young adults report more loneliness than people twice their age. Harvard’s Making Caring Common survey put loneliness among 18-to-25-year-olds at 61% against a national figure of 36%. Gallup’s 2023 global survey found 27% of adults aged 19 to 29 reporting loneliness worldwide, against 17% among those 65 and older. A 2026 eight-country study reported that nearly half of young adults across those countries described themselves as lonely, with the association to depression and anxiety holding after adjusting for demographics and prior diagnoses. Cigna’s 2025 survey found Gen Z and Millennial respondents reporting more loneliness than Gen X or Baby Boomers, despite reporting greater technological connectedness.

The pattern is consistent enough that it has stopped being treated as a finding and started being treated as a fact to explain. That is worth pausing on, because almost all of the surveys producing this pattern are cross-sectional. They sample different people at different ages at one point in time. What they cannot distinguish is whether young adults today are lonelier than young adults were a generation ago, or whether being young has always been lonelier than being old, and the surveys are simply the first to measure it consistently enough to notice.

Three explanations, one dataset

A cross-sectional snapshot showing that 27-year-olds report more loneliness than 70-year-olds is compatible with at least three distinct underlying processes, and a single wave of data cannot tell them apart.

The first is a cohort effect: something particular to people born in a given period – the timing of smartphone adoption, the disruption of early adulthood by the pandemic, changes in how young adults form households – has left this generation lonelier than prior generations were at the same age, and that elevated loneliness will likely persist as they get older.

The second is an age effect: loneliness has always been higher in the unsettled years of early adulthood, when people are leaving school, moving cities, and losing daily proximity to established friend groups, and it falls as people settle into stable partnerships, workplaces, and neighborhoods. On this account, today’s 27-year-olds are not carrying anything unusual; they are simply at the age where loneliness peaks, and will look more like today’s 70-year-olds by the time they reach that age themselves.

The third is a measurement effect: younger respondents may simply be more willing to endorse the word “lonely” on a survey, whether because of reduced stigma, greater familiarity with the vocabulary of mental health, or differences in how the item lands depending on how a life stage is culturally scripted. If that is what is happening, the age gradient in the data reflects reporting behavior rather than a real gradient in underlying social deprivation.

A cross-sectional survey cannot separate these three. Only a study that follows the same people as they age – a genuine longitudinal panel – can do that, because it can hold cohort membership fixed and watch the age effect unfold, or hold age fixed and compare across cohorts. The literature reviewed here has almost none of that for young adults. The one longitudinal example available, Japan’s JACSIS study, tracked isolation and loneliness prevalence in the general population across 2020 and 2021, which speaks to change during the pandemic but not to the shape of the age gradient over a life course.

Why this is not a pedantic distinction

The three explanations point to three different policy responses, and treating the cross-sectional finding as though it already answered the question risks funding the wrong one.

If the loneliness gradient by age is mostly a cohort effect tied to something distinctive about how this generation came of age, the sensible policy target is that specific mechanism. Jonathan Haidt’s The Anxious Generation argues for exactly this reading: that the shift from play-based to phone-based childhood drove a cohort-specific rise in loneliness and related conditions starting in the early 2010s, and that the fix is structural change to childhood and adolescence, particularly around smartphone and social media use. The companion evidence review that Haidt maintains with Jean Twenge compiles supporting studies, but also includes dissenting evidence – several researchers have argued the causal claim is stronger than the underlying correlational data can support. That disagreement is itself informative: it shows that even researchers working from the same body of studies disagree about whether a cohort-specific mechanism has been established, as opposed to merely observed.

If the gradient is mostly an age effect – loneliness peaking during the unsettled transitions of early adulthood regardless of when someone was born – the more durable response looks less like intervening on a generation and more like building infrastructure that eases those specific transitions: affordable ways to meet people after moving to a new city, workplace and campus structures that create recurring contact, and stable third places of the kind Ray Oldenburg described and Eric Klinenberg later argued has been eroding. A study of college students coming out of the pandemic found that the connections most associated with wellbeing were not simply more contact, but more regular and higher-quality contact – a finding that fits an age-transition account better than a generation-specific one, since it says something about what kind of connection helps at a particular life stage rather than about what happened to a particular cohort.

If the gradient is substantially a measurement effect, the policy implication is different again: it argues for caution in treating headline loneliness rates as evidence of a worsening crisis, and for investment in the kind of comparable, repeated-panel instrument that could separate reporting differences from real ones. The OECD’s 2025 comparison across member countries found that self-reported feelings of connection worsened only relatively recently, while in-person meeting frequency had been declining for longer – a divergence between behavior and self-report that is itself a warning against reading loneliness trends off subjective survey items alone.

The technology question sits inside the same gap

Much of the current debate about young-adult loneliness folds in a further variable – digital and AI-mediated substitution for in-person contact – and the same design problem recurs there in sharper form. A commentary from George Mason University’s public health faculty frames AI companionship as a substitution risk rather than a straightforward remedy, and a 2026 review of AI chatbots and digital companions found genuinely mixed evidence, with effects depending on user characteristics, chatbot design, and intensity of use. Reporting on this literature found both short-term comfort and long-term deepening of distress reported in different studies, with the distinction tracking duration of use rather than a simple presence-or-absence of the technology.

None of this evidence is longitudinal in the sense that matters here either. Almost all of it compares heavier users to lighter users at one point in time, which cannot establish whether heavy use causes later loneliness, whether already-lonely people gravitate toward heavier use, or both. Given that young adults are the heaviest adopters of most of these technologies, any cohort-versus-age ambiguity in the loneliness data is compounded by an equally unresolved direction-of-causation ambiguity in the technology data.

What would actually settle this

A study capable of separating these explanations would need to follow the same individuals across multiple life stages, using a consistent instrument, across at least two distinct birth cohorts, so that age and cohort could be statistically disentangled rather than confounded in a single cross-section. It would also need to pair the subjective loneliness measure with an independent, structural measure of network size or contact frequency, since isolation and loneliness are known to diverge, sometimes substantially, and a purely self-report panel would still leave the measurement-effect explanation on the table.

No such study exists yet for the current generation of young adults, at the scale or duration needed to answer the question. Until one does, the safest reading of the data is the narrowest one: young adults today report more loneliness than older adults do, on comparable instruments, across multiple countries and multiple years. Why that gradient exists, and whether it will still be there when today’s 25-year-olds turn 65, is not something the current evidence base is built to answer.