Feeling perfectly content with one’s own company may not erase every cognitive risk associated with having too little contact with other people. A large US analysis found that social isolation was linked to poorer later-life cognitive trajectories even when loneliness was held to its expected course. In the researchers’ model, loneliness explained only 6 percent of the estimated effect.

That finding does not mean lonely feelings are trivial, nor does it prove that scheduling more social events will prevent dementia. It makes a narrower and useful point: objective isolation and subjective loneliness overlap, but they are not interchangeable. A person can experience one without the other, and each may need a different response.

The study, published online in December 2025, was led by Jo Mhairi Hale, Angelo Lorenti and Solveig A. Cunningham. It used the US Health and Retirement Study, a nationally representative longitudinal survey of adults over 50 and their spouses.

The analysis included 30,421 people aged 50 to 94 and 137,653 person-waves, meaning observations collected from the same people at different survey waves. Participants needed to appear in at least two consecutive waves. The main analysis covered 2004 through 2018, when the relevant measures were collected consistently every two years.

Social isolation was treated as an objective condition. Loneliness was treated as a subjective appraisal: the gap between the social relationships someone has and those they want. The distinction is easy to blur in ordinary speech, but it was visible in the data. Only 55 percent of observations in which people reported loneliness were also classified as more isolated. About 26 percent of the more isolated observations involved people who said they had not felt lonely.

That is why the headline’s phrase “whether people felt lonely or not” needs a little care. The researchers did not simply divide participants into cheerful loners and unhappy loners. They modelled the part of isolation’s estimated effect that operated through a measured loneliness response and the much larger part that did not.

The team built an eight-point social isolation index. It included partnership status, time spent with family and friends, barriers to sociability, religious participation and hours spent volunteering. Scores from six to eight were classified as more isolated; scores of five or below as less isolated. Household composition was kept separate so the researchers could later model an intervention aimed at people living alone.

Loneliness came from a single item in the Center for Epidemiological Studies Depression scale: whether the respondent had felt lonely during the previous week. That yes-or-no measure was available at every wave, though it plainly cannot describe the frequency, intensity or meaning of loneliness as fully as a longer scale.

Cognition was scored from zero to 27 using immediate and delayed word recall, subtracting seven repeatedly from 100 and counting backward from 20. These tasks sample memory and mental control. They do not by themselves diagnose dementia, and a fraction of a point on the scale should not be translated into a diagnosis.

Earlier research has not always cleanly separated the social and emotional variables. A 2019 systematic review of social isolation and later-life cognition found broadly consistent associations, but also substantial variation in measures and study design. A later study of transitions into and out of isolation and loneliness likewise treated them as changing states rather than fixed personal traits.

To untangle a complicated feedback loop, the researchers used a method called the parametric g-formula. They first modelled the observed course of isolation, loneliness, cognition and other changing characteristics. They then simulated a policy-like shift in which people classified as more isolated moved into the less-isolated category.

The method allowed them to compare estimated cognitive trajectories under that hypothetical shift with the trajectories expected under the observed course. For the mediation step, loneliness was kept at its expected natural level so that the simulated isolation change could not alter it. The difference between the total estimated effect and that direct effect was counted as the portion operating through loneliness.

Loneliness accounted for 6 percent of the total estimated effect, with separate estimates of 8 percent for men and 5 percent for women. In other words, about 94 percent of the modelled relationship between isolation and cognition travelled through pathways other than the measured loneliness item.

It does not mean isolated participants lost 6 percent of their cognitive ability. It does not mean 6 percent became cognitively impaired. Nor does it show that loneliness matters for only 6 percent of people. Mediation percentages describe how an estimated effect is partitioned within a particular statistical model.

Across the age trajectory, the simulated reduction in isolation was associated with a cognitive score about 0.19 points higher on the 27-point scale, with a 95 percent confidence interval from 0.13 to 0.24. The estimated difference was 0.15 points for men and 0.21 for women.

The authors called the effect modest. For context, the average modelled cognitive score declined by roughly nine points between ages 50 and 94. The estimate also represented a plausible shift from the more-isolated category into the less-isolated category, not a magical move from severe isolation to constant companionship.

The protective association appeared across women and men, White, Black and Latinx participants, and educational groups, although the size varied slightly. A targeted simulation focused on isolated people living alone, who made up about one fifth of observations. That scenario captured around half of the population-wide estimated protective effect. It was a statistical intervention, not a programme that researchers actually delivered.

The result fits a broader pattern in which different social measures predict different outcomes. One longitudinal analysis found that isolation and loneliness showed distinct associations with later health and wellbeing. Science Blog has also covered evidence that smaller social circles in older age are not automatically worse ones. Network size, contact, relationship quality and loneliness answer different questions.

The new study did not identify the other 94 percent of pathways. That is an important boundary. Still, the distinction points researchers toward mechanisms that do not require emotional distress.

Conversation can exercise memory, attention and language. Social plans can prompt movement, travel and engagement with changing environments. Other people can notice health changes, offer practical help and reinforce routines. Volunteering, group membership and regular visits may provide cognitive stimulation and access to information even when their absence does not feel painful.

Those are plausible explanations, not findings from this mediation analysis. Physical activity, sleep, sensory loss, transport, illness and socioeconomic resources may also influence both social participation and cognition. The National Academies’ review of isolation and loneliness in older adults similarly treats the two as related but distinct public-health concerns with multiple possible pathways.

Loneliness remains important in its own right. It is associated with distress and numerous health outcomes, and an isolated person who feels fine today may not remain unaffected after bereavement, disability or a change in mobility. The study’s result argues against using loneliness as the sole screen for social disconnection, not against taking loneliness seriously.

The g-formula is more sophisticated than a simple correlation. It was chosen because isolation, loneliness, cognition and health can influence one another over time. The researchers adjusted for birth cohort, region, race and ethnicity, childhood socioeconomic conditions, education, wealth, partnership, household composition, employment, selected chronic conditions and self-reported depression.

Yet the authors explicitly note that the method does not guarantee a causal conclusion. Its interpretation depends on assumptions about correct model specification, consistency, sufficient overlap between groups and the absence of important unmeasured confounding. Cognitive decline can itself shrink a social world, so reverse causation cannot be made to disappear merely by using causal language.

The study also excluded participants who required a proxy respondent because the cognitive and loneliness questions depended on self-report. That may have removed people with more severe impairment and, the authors say, could bias the estimated relationship downward. Its isolation index could not capture every social domain, while the one-item loneliness measure compressed a complex experience into a binary answer. The data also ended in 2018, before the COVID-19 pandemic disrupted social life.

The most defensible conclusion is therefore not that company is a medicine or solitude a disease. It is that feeling lonely is an incomplete way to identify social isolation. Older adults who do not report loneliness can still have very little contact, participation or practical connection. If future trials confirm that rebuilding those opportunities protects cognition, programmes may need to address the structure of social life as well as the feeling of being alone.