What Would a Null Result on Loneliness Even Look Like

KFF’s June 1 survey of racism, discrimination, and health does something most national loneliness surveys do not: it breaks social support networks down by race and ethnicity and links network size to reported experiences of discrimination. That is a genuine contribution to a literature that has, until recently, treated loneliness as largely demographically flat. It is also a useful occasion to ask a question that rarely gets asked directly. In a body of research this consistent, what would a negative finding actually look like — and would it survive contact with the rest of the field?

The pattern that never breaks

Every major survey of the last decade lands in roughly the same place. The AARP Foundation’s 2018 survey of adults 45 and older, using the 20-item UCLA Loneliness Scale, found one in three lonely. Harvard’s Making Caring Common project put serious loneliness among all Americans at 36% in 2021, rising to 61% among young adults. Gallup’s 142-country survey in 2023 found 24% lonely worldwide. The U.S. Surgeon General’s 2023 advisory cited roughly half of American adults reporting loneliness. The instruments differ, the populations differ, the wording differs — and yet nobody publishes a survey finding that loneliness is rare, declining, or unrelated to health. The KFF survey does not break that pattern. It finds that people who report experiencing discrimination have smaller social support networks, which is consistent with, not contrary to, everything published before it.

This is not itself suspicious. Loneliness may simply be common and getting more so, and a real phenomenon should show up reliably across methods that measure it competently. But a literature that never produces disconfirming results should prompt a specific kind of scrutiny: not “is this true” but “what mechanism would produce a null result here, and has anyone built a study capable of finding one?”

Three ways a null result could arise, and why none of them show up

First, a null result on prevalence: a well-designed survey finds loneliness flat or falling. This essentially never happens in the sources available, and one reason is structural. Studies asking about loneliness are disproportionately funded and published because someone already suspects loneliness is a problem worth documenting. The BMC Public Health review of the field’s state in 2023 flagged inconsistent measurement as a barrier to comparison, but did not flag publication bias toward positive findings as a barrier to interpretation — a gap in the self-criticism of the field itself.

Second, a null result on the health association: a well-powered study finds no relationship between social isolation and mortality once other factors are controlled. Holt-Lunstad’s 2015 meta-analysis in Perspectives on Psychological Science came closest to testing this rigorously, and the association survived adjustment for health status, with isolation and loneliness independently predicting earlier death (OR 1.29 and 1.26 respectively). That the association survives serious adjustment is real evidence. But a meta-analysis of published studies is, by construction, a meta-analysis of studies that found something worth publishing. The absence of a well-known null result on this question is not the same as evidence that none exists.

Third, a null result on causation: an intervention trial finds that increasing social contact does not improve health outcomes. This is the gap the American Heart Association’s 2022 scientific statement named explicitly, describing the absence of intervention evidence as the central research need in the entire field. That absence cuts both ways. It means the causal claim connecting isolation to cardiovascular and brain health has not been falsified. It also means it has not been tested with the kind of design that could falsify it.

Why the KFF survey doesn’t change the calculus

None of this makes the KFF findings suspect. Smaller social support networks among people reporting discrimination is a plausible, mechanistically sensible finding, and it fills a real gap — most national surveys report loneliness by age and not by race. The point is narrower: this survey, like nearly every one before it, is cross-sectional. It cannot distinguish whether discrimination shrinks networks, whether people with smaller networks are more attentive to and more likely to report discrimination, or whether some third factor — residential segregation, income, health status — drives both. KFF does not claim otherwise; the report is descriptive, not causal, and treating it as more than that would be the analyst’s error rather than the survey’s.

A literature that wants to move from “loneliness is common and correlated with poor health” to “reducing loneliness improves health” needs a different kind of evidence than another prevalence estimate, however well disaggregated. It needs trials with a real chance of showing nothing — interventions delivered at scale, pre-registered outcomes, and researchers willing to report it if the intervention does not work. Until that evidence exists in volume, every new survey confirming the general shape of the problem, KFF’s included, adds precision to where loneliness concentrates without adding proof of what would relieve it.