Serious Loneliness and the Limits of a Single Survey

Making Caring Common, a project at the Harvard Graduate School of Education, published survey results on 9 February reporting that 36% of Americans feel “serious loneliness.” Among young adults aged 18 to 25 the figure was 61%, and among mothers with young children 51%. Forty-three percent of young adults reported that their loneliness had increased since the start of the pandemic.

Those numbers have moved quickly through the press, and they are worth taking seriously. They are also worth reading with more care than a headline permits.

What the survey measured

The phrase doing the work here is “serious loneliness.” It is not a diagnostic category, and it is not the same construct as the UCLA Loneliness Scale scores that most of the epidemiological literature rests on. Different instruments produce different prevalence estimates from the same population, and the gap between them is not small.

This is visible in the existing data. Cigna’s 2020 index, using its own instrument, reported that 61% of American adults sometimes or always feel lonely — a figure that sounds far larger than Harvard’s 36%, but which is measuring a broader and less severe state. Neither number is wrong. They are answers to different questions.

The practical consequence is that prevalence figures from different surveys should not be placed on the same axis, and a change between two surveys using different instruments is not evidence of a trend.

What the survey does not establish

The survey is cross-sectional. It reports what a sample of people said at one point in time. It cannot establish that loneliness rose because of the pandemic, only that 43% of young adults said it had — a self-reported retrospective judgement, collected during a period when respondents had ample reason to attribute any deterioration in their lives to the same cause.

This is not a criticism of the design. Retrospective self-report is often the only instrument available at speed during an unfolding event. It is a caution about what weight the finding can carry.

Why the direction of the finding is still credible

Two things make the broad picture more persuasive than any single survey.

The first is convergence. The National Academies concluded in early 2020, before the pandemic, that roughly a quarter of adults aged 65 and older were socially isolated, and recommended that health systems assess for it routinely. Cigna’s index had been reporting elevated figures for years. Harvard’s finding sits inside an existing pattern rather than announcing a new one.

The second is that the outcome literature is considerably stronger than the prevalence literature. Holt-Lunstad and colleagues’ 2015 meta-analytic review found odds ratios of 1.29 for social isolation, 1.26 for loneliness, and 1.32 for living alone in relation to early mortality, with effects that persisted after adjustment for health status. Their earlier 2010 review, across 148 studies and more than 300,000 participants, found that stronger social relationships were associated with a 50% increase in the likelihood of survival.

Those are pooled estimates across many studies and many populations. They do not depend on any one instrument, and they are what makes the prevalence question consequential in the first place.

The distinction that keeps getting lost

Social isolation and loneliness are not the same thing, and the survey’s framing does not always keep them apart. Isolation is a measurable property of a person’s social network: how many contacts, how often, how varied. Loneliness is the subjective experience of a gap between the relationships someone has and the ones they want. Cacioppo’s framing of loneliness as an aversive signal — closer to hunger than to a personality trait — is useful precisely because it separates the felt state from the structural one.

People can be isolated without being lonely, and lonely in the middle of a dense social network. The 2015 meta-analysis found both to be independently predictive of mortality, which means collapsing them into a single “epidemic” figure loses information that matters for anyone designing a response.

What would strengthen the picture

Three things, in order of tractability. Consistent instruments across major national surveys, so that figures can be compared rather than merely quoted alongside one another. Longitudinal panels rather than repeated cross-sections, so that change can be observed instead of recalled. And separate reporting of isolation and loneliness, so that interventions can be aimed at whichever one a given population actually has.

None of that diminishes the Harvard finding. It sharpens the question it raises.