Reading a Loneliness Trend Line: What Repeated Surveys of Young Adults Can and Cannot Show

Cigna’s 2020 loneliness report states that 61% of U.S. adults are lonely, “up 7 points” from the previous year. That second number is doing more work than the first. A prevalence figure is a description of one sample at one moment. A change figure is a claim that two samples, measured a year apart, differ in the population and not merely in how they were assembled and asked.

For young adults the stakes are high, because nearly every claim that loneliness is concentrated in the 18-25 range and worsening rests on a comparison of this kind. The same Cigna report puts loneliness at 73% among workers aged 18-22 and above 80% among employed members of Gen Z. Harvard’s Making Caring Common survey, published in February 2021, reports 61% of adults aged 18-25 experiencing serious loneliness. These are the two most quoted figures in the field. Neither is a trend line, and understanding why is a prerequisite for using them.

What comparability across waves actually requires

A repeated cross-sectional series supports a change estimate only if six things hold constant between waves.

The instrument: the same scale, with the same number of items, scored the same way. A 20-item UCLA Loneliness Scale and a three-item short form do not produce interchangeable prevalence figures, and neither is interchangeable with a single direct question about feeling lonely.

The wording and response options: “sometimes or always” collapses two response categories into a prevalence threshold. Moving that threshold, or altering the labels on the scale, changes the headline without changing anything in the population.

The mode: telephone, mailed paper, and online panel administration produce different answers to sensitive items. Loneliness is a sensitive item. Admitting to it involves a social cost that varies with whether a live interviewer is present.

The sample frame and denominator: who is eligible to be counted. The field window: whether both waves were collected in the same season and under comparable external conditions. And the weighting scheme, which determines how the raw responses are projected onto the population.

Change one of these and the resulting difference is a mixture of real movement and artefact, in unknown proportions. Change several and the comparison is not recoverable.

The denominator problem in the workplace series

Cigna’s report is titled Loneliness and the Workplace. Its youngest-cohort figures are stated for workers aged 18-22 and for employed Gen Z. That framing matters more for this age band than for any other, because a substantial share of 18-22 year olds in the United States are full-time students, and their employment status is unstable across a year in a way that a 45-year-old’s is not.

If the eligible population shifts from all adults toward employed adults, or the share of the young cohort in work moves, the young-adult figure moves with it — independent of anyone becoming lonelier. A 73% figure for young workers and a hypothetical figure for all young adults are not the same statistic, and a year-over-year comparison that quietly slides between them is not measuring change in loneliness.

The other consequence of the workplace frame is that the same series produces the report’s most policy-legible findings — lonely workers missing work twice as often through illness and five times as often through stress — from a sample selected on employment. Those absenteeism estimates are informative about workplaces. They cannot be read back onto young adults generally.

Retrospective change is not a measured change

Making Caring Common reports that 43% of young adults said their loneliness had increased since the pandemic began. This is frequently cited as evidence of a rise. It is evidence of something else: a retrospective self-assessment of one’s own trajectory, collected at a single point in time.

Retrospective change items and two-wave measurement can disagree substantially, and when they do, the retrospective item is the less reliable of the two. Respondents are asked to compare their present state with a remembered state, under conditions where the present state colours the recollection. The item is also asked against a highly salient public narrative about pandemic loneliness, which supplies its own answer.

The 36% overall and 61% young-adult serious-loneliness figures from that survey stand on their own as prevalence estimates, and the survey’s most striking finding does not depend on any trend claim at all: roughly half of lonely young adults said no one had taken more than a few minutes in recent weeks to ask how they were doing in a way that felt genuine. That is a cross-sectional observation, and it is the stronger part of the report.

The thirty-year comparison

The Survey Center on American Life’s 2021 friendship survey reports that 12% of Americans have no close friends, against 3% in 1990, and that the share of men with at least six close friends fell from 55% to 27% over the same span. The 1990 benchmark was not collected as an earlier wave of the same series. It comes from a different survey, run by different people, under field conditions that cannot now be reconstructed and matched.

This does not make the comparison worthless. A fourfold change is large enough that it is implausibly all artefact, and the male-specific pattern — a fivefold rise in men reporting no close friends — is internally consistent across several items. But the honest statement is that the direction is credible and the magnitude is not estimable. Anyone quoting “3% to 12%” as a measured trend is quoting a cross-series comparison as though it were a within-series one.

By contrast, the AARP Foundation’s 2018 survey of 3,020 adults aged 45 and older used the 20-item UCLA scale, which makes its one-in-three prevalence figure directly comparable to the academic literature — and entirely unusable as an anchor for young adults, who fall outside its frame.

Changing the construct is worse than changing the instrument

A series can also break by measuring a different thing. Loneliness is a subjective state — Cacioppo’s framing of it as an aversive signal that motivates reconnection is a claim about internal experience. Social isolation is a structural property of a network, and the National Academies’ estimate that roughly a quarter of adults over 65 are socially isolated is a count of ties and contacts, not of feelings.

Holt-Lunstad’s 2015 meta-analysis found both to be independently predictive of early mortality, with odds ratios of 1.26 for loneliness, 1.29 for isolation, and 1.32 for living alone, and reported the association was stronger in samples averaging under 65. Independent prediction is precisely why the two cannot be substituted for one another mid-series. A wave that swaps a loneliness item for a network-size item has not refined its measurement; it has started answering a different question.

What a usable series would look like

The UK’s 2018 strategy did the one structurally important thing here: it embedded standard loneliness questions in the Office for National Statistics’ continuing surveys, so that the instrument, mode, and frame are held fixed by an institution whose business is comparability rather than publication.

A U.S. equivalent would fix the wording in advance, keep the mode constant or run a bridge sample when it changes, oversample 18-25 year olds so cohort estimates carry usable confidence intervals, publish item-level distributions rather than a single collapsed prevalence threshold, and report the field dates as prominently as the publication date. Holt-Lunstad’s 2021 argument for treating connection as a modifiable prevention target alongside diet and smoking depends on it. Prevention frameworks need a baseline that can be moved and measured. At present, the young-adult figures in circulation cannot tell whether anything has moved at all.