What a Null Result in Loneliness Research Would Actually Look Like

Cigna reported in 2020 that 61% of U.S. adults sometimes or always felt lonely. Cigna reported in 2025 that the figure was 57%. Harvard’s Making Caring Common project put serious loneliness at 36% in 2021. Gallup, surveying 142 countries, found 24% feeling very or fairly lonely in 2023. Four credible organizations, four different numbers, none of them wrong exactly, none of them comparable. This is not a story about loneliness rising or falling. It is a story about what happens to a claim when almost nobody in the field is set up to falsify it.

The literature on social connection is unusually consistent in one respect: it almost always finds something. Meta-analyses find elevated mortality risk. Systematic reviews of social prescribing find improved wellbeing. National surveys find substantial loneliness. It is worth asking, as a matter of method rather than politics, what a negative or null result would look like in each of these areas, and why so few of the published studies are actually built to produce one.

Three different claims that could each be null

“Loneliness is a growing epidemic,” “loneliness is bad for health,” and “programs that increase social contact reduce loneliness” are three separate empirical claims, and a null result would look different for each.

A null result on the epidemic claim would require the same instrument, applied to the same population, at multiple points in time, showing no change. That is a higher bar than it sounds. Most of the widely cited figures are single cross-sectional snapshots using different question wording, different response scales, and often bespoke items rather than a validated instrument. The AARP Foundation’s 2018 survey of adults 45 and older used the 20-item UCLA Loneliness Scale, which puts it on stable methodological ground; a great deal of the Cigna and Gallup work uses shorter, study-specific items that cannot be assumed to measure the same construct. Comparing Cigna’s 61% to Harvard’s 36% to Gallup’s 24% is not tracking a trend. It is comparing three different rulers. A genuine test of whether loneliness is rising would need a repeated-measures design on a fixed instrument — something closer to the JACSIS study in Japan, which followed the same cohort with the same measure across 2020 and 2021 specifically so that change over time could be attributed to something other than instrument drift. That kind of design is rare, and it is the only kind capable of returning a credible null on the trend question.

A null result on the health-effects claim would look different again. It would mean that after adequate adjustment for confounding — baseline health, depression, socioeconomic status, health behaviors — the association between isolation or loneliness and mortality shrinks toward zero. This has not happened. Holt-Lunstad’s 2010 meta-analysis, covering 148 studies and more than 308,000 participants, found weak social relationships associated with a 50% increase in mortality risk, an effect comparable to well-established risk factors. The 2015 follow-up, disaggregating isolation, loneliness, and living alone, found odds ratios of 1.29, 1.26, and 1.32 respectively, and reported that these effects survived adjustment for health status. The American Heart Association’s 2022 scientific statement reached broadly similar figures — roughly 29% increased risk of heart attack or death from heart disease, 32% increased risk of stroke — again after adjustment. These are not fragile results that a single reanalysis would erase. If a null result exists here, it has not yet been published in a form that has entered the mainstream literature, and it is worth asking whether that absence itself is informative.

The intervention gap is where a null result is most plausible — and least tested

The AHA statement is unusually candid about where the evidence actually runs out. It identifies the absence of intervention evidence as the central research gap: the observational link between isolation and cardiovascular and brain health is well established, but whether intervening on isolation changes outcomes is largely untested. This is the area where a genuine null result is most likely to eventually surface, because it is the area furthest from being settled and the one with the fewest well-controlled studies.

Consider the social prescribing literature, which sits at the center of policy enthusiasm for connection-based interventions. A 2021 systematic review in Perspectives in Public Health found that all nine included studies reported positive individual impacts. A separate 2021 review in the International Journal of Environmental Research and Public Health reported improvements in self-esteem and confidence but explicitly flagged limited trial evidence and heterogeneity across programs. A 2022 qualitative meta-synthesis found participants describing benefits extending to restored purpose and participation. Taken together, this is a literature in which every study finds a positive effect. That should prompt more caution, not less. Uncontrolled pre-post designs are structurally prone to finding improvement regardless of the intervention’s actual effect, through regression to the mean, demand characteristics, and selection into programs by people already inclined to benefit. A literature that never returns a null result is not strong evidence that the intervention works; it is often evidence that the designs used cannot detect a null result even when one exists.

This is what makes the Lancet Healthy Longevity trial of volunteering among lonely older adults in Hong Kong worth naming specifically. It is a randomised controlled trial with a genuine control condition, which the source list flags as one of the few such designs in this literature rather than an uncontrolled programme evaluation. Whatever its result, the design itself is the point: an RCT is structurally capable of returning “no difference between arms,” in a way that a before-and-after survey of program participants is not. The scarcity of designs like it, set against the abundance of uniformly positive uncontrolled evaluations, is itself a finding about the state of the field. A 2023 review in BMC Public Health, mapping the current state of loneliness and isolation research broadly, identifies inconsistent measurement as a structural barrier to comparing results across studies — which is another way of saying that much of this literature is not yet built to contradict itself.

Isolation and loneliness may not even move together

One more place a partial null shows up: the assumption that isolation and loneliness track each other, so that reducing one reduces the other. A 2024 study in Scientific Reports examining their interplay by age found the relationship between the two constructs varies with age rather than holding constant — a structural property of a network does not map cleanly onto a subjective state at every life stage. That finding does not overturn either construct’s independent link to health outcomes, both well supported elsewhere. But it complicates any intervention logic that assumes fixing contact frequency will automatically fix the felt experience of loneliness, or vice versa.

A stronger evidence base here would look less like another prevalence survey and more like three specific things: repeated-measures studies on a fixed, validated instrument rather than cross-sectional snapshots; adjustment models in the mortality literature that are pre-registered to test whether the association actually can be explained away, rather than confirming that it cannot; and more randomised trials willing to publish a null result on the same terms as a positive one. Until that last piece exists in volume, it will remain hard to tell whether social prescribing and similar programs work, or whether the literature simply has not yet been built in a way that could tell the difference.