What a Longitudinal Study Would Show That a Snapshot Cannot

The Harvard Making Caring Common survey found that 61% of young adults aged 18 to 25 reported serious loneliness in early 2021, compared with 36% of Americans overall. Cigna’s 2020 workplace report put the figure for the youngest workers, aged 18 to 22, at 73%. These numbers get repeated often, and they are startling. What they cannot do, on their own, is tell anyone whether a given 22-year-old who is lonely this year will still be lonely at 30, or whether loneliness at 22 is a cause of anything that happens afterward, or simply a marker of a phase most people pass through.

That distinction sounds pedantic. It is not. It determines what kind of policy or clinical response makes sense, and almost every widely cited figure on youth loneliness comes from a design that cannot answer it.

What a cross-sectional survey actually captures

A cross-sectional study asks a sample of people how they are doing, once, at a single point in time. The Harvard survey, the Cigna report, and the AEI survey on American friendship are all of this kind: each drew a sample, asked a set of questions, and produced a snapshot. The AEI survey found that 15% of men reported no close friends in 2021, up from 3% in 1990 — but that “1990” figure comes from a different survey of different people, not from following the same men for three decades. It is a comparison of two snapshots taken 31 years apart, not a trajectory.

This matters for three reasons.

First, a snapshot cannot distinguish a cohort effect from an age effect. If young adults report more loneliness than older adults, as the Harvard survey found, that could mean loneliness declines as a person ages — or it could mean people born more recently are lonelier throughout their lives than people born earlier, at every age, for reasons specific to their generation. Those are different phenomena requiring different responses, and a single survey wave cannot tell them apart, because it never observes the same people at two ages.

Second, a snapshot cannot establish sequence. The Harvard survey reported that about half of lonely young adults said no one had checked in on them meaningfully in recent weeks. It is tempting to read this as cause: thin support networks produce loneliness. It is equally consistent with the reverse: loneliness makes people withdraw, so they stop reaching out and stop being reached, and the sparse contact is a symptom, not a cause. Cross-sectional data cannot order those events in time, because it was collected at one moment. Holt-Lunstad’s 2015 meta-analysis, drawing largely on studies designed to track mortality risk over years, was able to establish that isolation and loneliness predicted earlier death even after adjusting for baseline health — but that kind of claim requires a design that observes people before the outcome occurs, not one that asks everyone the same question on the same day.

Third, a snapshot conflates state and trait. Some fraction of any lonely-at-one-point-in-time group is passing through a transient rough patch — a breakup, a move, a job loss — and will look entirely different a year later. Another fraction is chronically lonely and will still be lonely at the next measurement. A single wave of data cannot separate these groups; it reports 61% lonely without any way of knowing how much of that 61% is temporary. This is the single largest thing a repeated-measures design would add for young adults specifically, since early adulthood is a period of unusually high residential, relational, and occupational turnover, and unusually high rates of transient distress that resolve on their own.

What following the same people would show

A longitudinal design recruits a cohort and returns to the same individuals at intervals — a year later, five years later, longer. Applied to young-adult loneliness, this would let researchers answer questions the existing survey literature cannot.

It would show what proportion of loneliness at 20 resolves by 25 without intervention, versus what proportion persists. That single number would reframe the policy conversation. If most youth loneliness is transient, resources belong in low-cost, low-friction support during known transition points — leaving home, finishing school, changing jobs. If a substantial minority is persistent, that argues for something closer to clinical screening, of the kind the National Academies has proposed for older adults through primary care.

It would also let researchers test what predicts persistence rather than just prevalence. The AARP survey of adults 45 and older found that network size and diversity, along with physical isolation, were the strongest predictors of loneliness in that population, and that people who spoke with neighbours were far less likely to be lonely than those who never did. Those are cross-sectional associations within an older cohort. Whether the same factors predict who stays lonely among young adults — as opposed to who happens to be lonely on the day of the survey — is a distinct empirical question, and one that requires watching the same people’s networks and loneliness scores move together over time.

Finally, a longitudinal design applied across the pandemic period specifically would let researchers separate a genuine pandemic effect from a pre-existing trend that the pandemic merely made visible. The AEI data suggest friendship networks had been thinning for three decades before 2020. The Harvard survey found 43% of young adults reported increased loneliness since the pandemic began — a comparison the respondents made themselves, retrospectively, rather than one derived from measuring the same people before and after. Self-reported change is not the same evidence as two independent measurements of the same person at two points in time, because retrospective recall is itself distorted by current mood and by narrative expectation. A cohort recruited before 2020 and re-surveyed afterward would settle the question with actual before-and-after data rather than recollection.

Why this design is rare here

Longitudinal cohort studies are expensive, slow, and vulnerable to attrition — the people most likely to drop out of a multi-year study are often the same people who are hardest to reach because they are isolated, which can bias the very estimate the study is trying to produce. Loneliness research on younger cohorts has instead leaned on repeated cross-sections: the same questions asked of a fresh sample every year or two, which shows aggregate trends over time but never tracks an individual trajectory. That is a legitimate and useful design for tracking population-level change, and it is what underlies most of the youth loneliness figures now in circulation. It is a different thing from a cohort study, and the two get treated as interchangeable more often than the underlying data license.

None of this means the existing figures are wrong. A 61% loneliness rate among 18-to-25-year-olds in early 2021 is a real finding about a real sample at a real moment. It is a poor basis, on its own, for claims about causation, duration, or which young adults will still be lonely in five years. Those claims require watching the same people long enough to find out.