When Loneliness Gets a Fourth Category

A study published in the American Journal of Preventive Medicine in November 2025 reports that roughly 1 in 7 U.S. adults experience severe loneliness and 1 in 5 experience moderate loneliness, for a combined moderate-to-severe rate above 37%. The underlying data come from the 2022 Health Information National Trends Survey (HINTS-6), so the figures describe conditions from three years before the paper’s publication date. That gap matters less here than something else: the analysis did not simply count lonely versus not-lonely. It split respondents into discrete loneliness states — none, mild, moderate, severe — and that categorical structure is the whole reason the headline number looks the way it does.

This is worth stopping on, because most coverage of loneliness prevalence treats a single number as if it dropped out of the sky fully formed. It didn’t. Every prevalence figure is downstream of a decision about what counts as “lonely” and how finely to slice the space between “not at all” and “constantly.” When that decision changes, the number changes, even if nothing about the underlying population does.

What actually changed

The conventional approach in most of the survey literature is binary or near-binary: ask a person to rate their loneliness, then draw a line somewhere and report the share above it. Gallup’s global work asks whether someone feels lonely “a lot of the previous day” and reports a share who say yes — 24% worldwide in the 2023 analysis. The Cigna Group’s 2025 report asks about feeling lonely “sometimes or always” and lands at 57% of Americans, with 52% of workers specifically. Harvard’s Making Caring Common project asked about “serious loneliness” and got 36% nationally, with 61% among adults 18 to 25. Three different thresholds, three different instruments, three numbers that cannot be stacked next to each other and read as a trend.

The HINTS-6 analysis does something structurally different. Rather than asking whether a respondent clears a single bar, it assigns each person to one of four ordered categories. That is a meaningfully different measurement object. A binary instrument answers “what share of people are lonely.” A four-state instrument answers “how lonely, distributed across how many people,” and then lets the analyst choose where to draw the line for reporting purposes — moderate-to-severe combined, severe alone, or the full distribution. The headline figure of “over 37% moderate-to-severe” is a choice about aggregation, not a fixed fact about the population. A different aggregation choice, severe alone, produces roughly 1 in 7 — a number closer to what other surveys using stricter thresholds report.

Why this is not a small technical point

The Center for Social Connection has noted before, drawing on the BMC Public Health review of the field published in 2023, that inconsistent measurement is one of the two or three largest obstacles to synthesizing loneliness research. The HINTS-6 case shows exactly how that inconsistency compounds. A journalist or policymaker who wants to say “loneliness has gotten worse” or “loneliness affects more than a third of Americans” now has to choose which HINTS-6 aggregation to cite, decide whether it is comparable to Cigna’s “sometimes or always” figure, and decide whether the CDC’s own 2024 surveillance report on 2022 data — which used its own set of items — belongs in the same sentence. None of these decisions is obviously wrong. All of them produce different-looking claims from data that, in the HINTS-6 case, describe the same year.

This is the instrument mismatch problem in its purest form: not two surveys disagreeing about a genuine trend, but the same underlying construct, measured with different granularity, producing headline numbers that read as contradictory to anyone who does not look at the method section.

The case that gets this right

There is a useful contrast close at hand. AARP surveyed adults 45 and older on loneliness in 2018 and again in its 2025 follow-up report, Disconnected, using the same 20-item UCLA Loneliness Scale both times. That consistency is unusual enough in this field that it is worth naming as a virtue on its own. Because the instrument did not change, a shift in the reported rate between 2018 and 2025 can be read as an actual change in the population rather than an artifact of how the question was asked or how the response categories were collapsed. The 2018 survey, drawing on 3,020 midlife and older adults, found that roughly one in three adults 45 and older were lonely, and identified network size and diversity, along with physical isolation, as the strongest predictors — not age itself. Whatever the 2025 follow-up shows about direction of change, it shows it against a fixed baseline. That is rarer than it should be.

The National Academies’ 2020 consensus report on older adults draws the isolation/loneliness distinction explicitly, treating isolation as a structural property of a person’s network — how many contacts, how often, what kind — and loneliness as the subjective gap between desired and actual connection. HINTS-6 measures loneliness, not isolation, and the severity-state design is entirely about the subjective side. None of the four categories say anything about network size. A respondent classified “severe” could have a large network and feel disconnected from it anyway; a respondent classified “none” could have a genuinely thin network and simply not feel the absence. The categorical refinement improves resolution on one axis while leaving the other axis untouched, which is worth stating plainly rather than assuming the more detailed number is simply “better” than what came before.

What would actually help

Three fixes would make this literature easier to use, none of them exotic. First, when a study reports loneliness as a graded or categorical measure, the paper should also report the binary equivalent using a stated, standard cutoff, so it can be cross-walked against studies still using single-item measures. Second, repeated administrations of the same instrument to the same or comparable populations — the AARP model — should be treated as the gold standard for claims about trend, and single cross-sectional snapshots should not be used to argue that loneliness is “rising” or “falling” without that caveat attached. Third, isolation and loneliness need to stop sharing space in summary sentences as though they were the same variable measured twice. The HINTS-6 paper is a loneliness study. Whatever it finds says nothing directly about how isolated its severe-category respondents actually are, and treating the two as interchangeable is exactly the kind of conflation that makes the next round of headline numbers just as hard to reconcile as this one.